A Little Off » Code

Decoding DVD Subtitles with Golang

bemasher — Mon, 21 Nov 2011 06:38:44 +0000

I've always been very fond of subtitles but I'm not sure of the reason why. When transcoding my DVD's to play them on my network media player I realized I needed a good way to keep the subtitles without burning them into the video. The MPEG-4 container will happily include VOBSUB and SRT subtitle streams and my network media player handles this nicely.

The problem though is that including the VOBSUB's exactly as they appeared on the DVD is somewhat problematic, they're almost never the same style between movies, and sometimes they're just plain difficult to read. Converting them to SRT involves a fairly lengthy process of going through and indicating to an OCR program what each character is as it reads all of the subtitles and writes an SRT file. This is also difficult to correct if you mess up one character in the process of encoding.

So now that I've got the itch I decided to scratch it. I decided to write my own subtitle decoder that would write subtitles to images and a pseudo-OCR program to convert those images into individual character files. From there it would be fairly easy to write a quick interface that presents you with a list of letters at which point you can just fill in the character for each one. Once you've done this you can export it as your favorite text subtitle format in one shot instead of doing it as you go along.

The language I decided to write it in is Golang as I've been learning it for a few weeks now and It's currently my favorite language for a large number of reasons I won't get into here.

The first major challenge I ran into is that there's not really any standardized information about decoding DVD subtitles. I did find maybe 3-4 sites that have varying levels of detail into decoding DVD subtitles but there were still a lot of gaps in the information.

To start with, we need to decode MPEG Program Stream packets (PS), these contain MPEG Packetized Elementary Stream packets (PES). The PS header doesn't contain any information we need to decode subtitles. The PES header contains size of the packet's payload, offset to the payload and the length of the additional headers. SubStream refers to the stream id of the subtitle we're decoding. DataSize is the size of the subtitle payload. ControlPtr is the offset to the control sequences for describing the subtitle's payload.

1
2
3
4
5
6
7
8
9
10

type Packet struct {
PSHeader [14]uint8
PESHeader [4]uint8
PacketSize uint16
Extension uint16
HeaderSize uint8
SubStream uint8
DataSize uint16
ControlPtr uint16
}

To read data into this structure I've written the following method:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

func (p *Packet) Read(r io.Reader) {
binary.Read(r, binary.BigEndian, &p.PSHeader)
binary.Read(r, binary.BigEndian, &p.PESHeader)
binary.Read(r, binary.BigEndian, &p.PacketSize)
binary.Read(r, binary.BigEndian, &p.Extension)
binary.Read(r, binary.BigEndian, &p.HeaderSize)
r.(io.ReadSeeker).Seek(int64(p.HeaderSize), os.SEEK_CUR)
binary.Read(r, binary.BigEndian, &p.SubStream)
binary.Read(r, binary.BigEndian, &p.DataSize)
binary.Read(r, binary.BigEndian, &p.ControlPtr)

p.PacketSize -= uint16(p.HeaderSize) + 4

// Back up; DataSize and ControlPtr are part of the payload
r.(io.ReadSeeker).Seek(-4, os.SEEK_CUR)
}

We read each of the structure's fields in order. We skip the additional headers of the PES packet since we don't care about the data in it. We also compensate for the given packet size since we went ahead and read the SubStream and DataSize seperately. Before leaving this function we back up so that the file cursor is at the right position to start reading data from the offsets.

Subtitles may span more than one packet so we need to be sure to read packets until we've read the entire length of the subtitle given by Packet.DataSize.

1
2
3
4
5
6
7
8
9
10
11
12
13
14

func ReadSubtitle(s *os.File) (head Packet, data bytes.Buffer) {
for i := 0; ; i++ {
var pack Packet
pack.Read(s)
if i == 0 {
head = pack
}
ReadFrom(s, &data, int64(pack.PacketSize))
if data.Len() == int(head.DataSize) {
break
}
}
return
}

Now that the headers and information like payload size and offsets have been read we can start to decode the subtitle. The first things we need to decode are the control sequences. These sequences give us information about how long to display the current subtitle, it's color and other information like offsets to even and odd fields since the image data is interlaced.

1
2
3
4

type ControlHeader struct {
Date uint16
Next uint16
}

ControlHeader represents the start time and offset to the next control sequence. Once we've got this information we can read the controls.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

func ReadControlSequences(head Packet, data *bytes.Buffer) (rect Rect, payload Payload, even, odd uint16) {
payload.Read(data, head)

for {
var header ControlHeader
err := ReadInto(&payload.Control, &header)
if err != nil {
break
}
fmt.Printf("%+v\n", header)
end := false
for !end {
cmd, err := payload.Control.ReadByte()
if err != nil {
break
}
switch cmd {
case 0x00: fmt.Println("\tForced")
case 0x01: fmt.Printf("\tStart:\t\t%dms\n", 1024 * header.Date / 90)
case 0x02: fmt.Printf("\tStop:\t\t%dms\n", 1024 * header.Date / 90)
case 0x03:
fmt.Printf("\tPalette:\t%04X\n", payload.Control.Next(2))
case 0x04:
fmt.Printf("\tAlpha:\t\t%X\n", payload.Control.Next(2))
case 0x05:
buf := payload.Control.Next(6)
rect = Rect{((uint16(buf[1]) & 0xF) << 8) | uint16(buf[2]) - (uint16(buf[0]) << 4) | (uint16(buf[1]) >> 4) + 1, ((uint16(buf[4]) & 0xF) << 8) | uint16(buf[5]) - (uint16(buf[3]) << 4) | (uint16(buf[4]) >> 4) + 1}
fmt.Printf("\tDimensions:\t%+v\n", rect)
case 0x06:
buf := payload.Control.Next(4)
even = uint16(buf[0]) << 8 | uint16(buf[1])
odd = uint16(buf[2]) << 8 | uint16(buf[3])
fmt.Printf("\tOffsets:\t%d, %d\n", even, odd)
fmt.Printf("\tField Len:\t%d, %d\n", odd - even, uint16(payload.Data.Len()) - odd)
case 0xFF:
end = true
}
}
}
return
}

The control command is 1 byte and is followed by any parameters necessary for that particular control. The different controls are described below:

0x00 - Forced: subtitle displayed whether or not subtitles are selected//enabled. This is typically used for foreign language segments. Takes no arguments.
0x01 - Start: The time at which to start displaying the subtitle, this uses the Date field of ControlHeader. The time in milliseconds to start displaying the subtitle is given by the function: 1024 * ControlHeader.Date / 90. Takes no arguments.
0x02 - Stop: The time at which to stop displaying the subtitle, takes no arguments.
0x03 - Palette: Defines the four colors used for the subtitle. I've decided to ignore implementing this as I will be converting the subtitles to text. Takes 2 bytes of arguments, each color is one nibble.
0x04 - Alpha: Alpha channel information, determines which colors are opaque and which are transparent. Useful for determining the main color as the background will likely have complete transparency. Takes 2 bytes of arguments, each alpha is one nibble.
0x05 - Dimensions: Gives the dimensions of the subtitle image. Takes 6 bytes of arguments, each dimension value is 3 nibbles. Dimensions in pixels is given by the equation: (X1 - X0 + 1) x (Y1 - Y0 + 1)
- 0x*** - X0: Left-most x-axis bound.
- 0x*** - X1: Right-most x-axis bound.
- 0x*** - Y0: Top-most y-axis bound.
- 0x*** - Y1: Bottom-most y-axis bound.
0x06 - Field Offsets: Gives the offsets to the even and odd fields of the image. Takes 4 bytes of arguments, the first byte is the even field offset, the second byte the odd field offset. This will be useful for rendering each field line in the proper order.
0xFF - End Control: Signals the end of a control sequence.

Now that we've got some information about the dimensions and locations of the subtitle image we can look at decoding and drawing it. Subtitle images are run-length-encoded (RLE). The basic idea behind RLE is to compress the image data into a pixel color and a number of pixels to draw in that color. Using the format for subtitles each pixel is defined by the following alphabet where * represents a wildcard nibble:

0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xA, 0xB, 0xC, 0xD, 0xE, 0xF
0x1*, 0x2*, 0x3*
0x04*, 0x05*, 0x06*, 0x07*, 0x08*, 0x09*, 0x0A*, 0x0B*, 0x0C*, 0x0D*, 0x0E*, 0x0F*
0x01**, 0x02**, 0x03**
0x000*

To determine the color and number of pixels to draw we need to do a little bitwise arithmatic. The number of pixels to draw is given by the operation: X >> 2. The color is given by X & 0x03.

There is one character in the alphabet which has a special meaning and neither of the above operations apply to it. That is 0x000* which is a sort of carriage return character. It means simply fill the rest of the line with the given color. After every carriage return we need to read a line from the opposite field and reset the x position in the image to 0 and increment the y position.

Before we get into the code about drawing images I should mention one of the problems I ran into while writing this. The problem is that Golang doesn't provide any mechanism for reading nibble-aligned data. So I went ahead and wrote a small structure and a few methods for accomplishing this.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

type Nibbler struct {
r *bytes.Buffer
Current uint8
Aligned uint8
}

func NewNibbler(r *bytes.Buffer) Nibbler {
return Nibbler{r, 0, 0}
}

func (n *Nibbler) GetNibble() (b uint8, err os.Error) {
if n.Aligned == 0 {
err = ReadInto(n.r, &n.Current)
if err != nil {
return 0, err
}
}
n.Aligned ^= 4
b = (n.Current >> n.Aligned) & 0x0F
return b, err
}

The basic functionality is achieved by using some bitwise operations to switch which nibble we return each time the GetNibble method is called and reading a new byte every time we've read the 2nd nibble of the current byte. Access is provided to the Aligned field to determine if we're byte-aligned or not since we need to use this in the function that draws the subtitle images.

The following code decodes the RLE image and draws all of the pixels to an image of dimensions specified in the control sequence.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73

func DrawPixels(s *image.Gray, x uint16, y uint16, n uint16, c uint8) {
for i := 0; i < int(n); i++ {
s.SetGray(int(x) + i, int(y), image.GrayColor{(c + 1) << 6})
}
}

func ReadRLEImage(rect Rect, payload *Payload, even, odd uint16) (*image.Gray) {
subImg := image.NewGray(int(rect.w), int(rect.h))
bData := payload.Data.Bytes()
evenNibbler := NewNibbler(bytes.NewBuffer(bData[even:odd]))
oddNibbler := NewNibbler(bytes.NewBuffer(bData[odd:]))

var x, y uint16
done := false
field := true

for !done {
var b uint16
var t uint8

var currentNibbler *Nibbler

if field {
currentNibbler = &evenNibbler
} else {
currentNibbler = &oddNibbler
}

t, _ = currentNibbler.GetNibble()
b = (b << 4) | uint16(t)
if b >= 0x4 {
run := b >> 2
DrawPixels(subImg, x, y, run, uint8(b & 0x3))
x += run
} else {
t, _ := currentNibbler.GetNibble()
b = (b << 4) | uint16(t)
if b >= 0x10 {
run := b >> 2
DrawPixels(subImg, x, y, run, uint8(b & 0x3))
x += run
} else {
t, _ := currentNibbler.GetNibble()
b = (b << 4) | uint16(t)
if b >= 0x40 {
run := b >> 2
DrawPixels(subImg, x, y, run, uint8(b & 0x3))
x += run
} else {
t, _ := currentNibbler.GetNibble()
b = (b << 4) | uint16(t)
if b >= 0x100 {
run := b >> 2
DrawPixels(subImg, x, y, run, uint8(b & 0x3))
x += run
} else {
DrawPixels(subImg, x, y, rect.w - x, uint8(b & 0x3))
x = 0
y += 1
field = !field
if y >= rect.h {
done = true
}
if currentNibbler.Aligned != 0 {
currentNibbler.GetNibble()
}
}
}
}
}
}
return subImg
}

You'll notice I used the even and odd field offsets to create buffers for both the even and odd fields of the image. Then to switch between them, the pointer currentNibbler is switched between each field whenever we encounter a carriage return. I've also done some basic math in the DrawPixels function to evenly space the colors used in the subtitle throughout the greyscale range from 0 to 128.

The next step for this project is to write a program which can detect and separate images of each character from a subtitle image. After this I'll write a user interface for the user to give character meanings to each character image. From that an SRT file can be written using this character matrix. This is the same basic operation of most VOBSUB to SRT converters except that I aim to make it easier to use.

The complete source for this program can be found at: Gist: 1381809. Note that this program will read and decode only the first subtitle in the subtitle file. More work will be done on this when I've got time to make a more automated version that will read and decode all subtitles from a file. At some point in the future when I find that GeSHi supports Golang syntax highlighting, I'll update this post to make it more readable.

Newegg’s JSON API

bemasher — Thu, 17 Mar 2011 05:11:22 +0000

For the longest time I've wanted access to Newegg's product list. For me they've been one of the better and more structured websites for buying computer hardware. So naturally they're usually my first choice when it comes to finding a good deal on a particular piece of hardware. They're also rather useful for seeing what's out there since their product catalog is fairly complete.

A while back I had started wanting to sort through items to heuristically pick the best deal based on a number of features Newegg generally provides for each item. This method works pretty well on SSD's and system memory. But until a recent discovery I was limited to scraping Newegg's website in order to get any kind of information from them. If you've ever tried this sort of thing you know that it is messy and generally a bad idea because every single time Newegg changes the structure of their website or any minute detail this will almost always break your scraping script.

The discovery came in the form of a mobile application for Android^[1]. The mobile app lets you browse their website in a clean and fast manner. But what got me thinking is that unlike some other mobile applications out there that are just application wrappers for the mobile version of their websites this one operates directly through the native GUI. Now this is where it got interesting. I knew that if Newegg had written the app to use the native GUI then they had to be providing the data to it somehow and I knew it had to be more structured than HTML scraping like what I've been doing^[2]. You have no idea how happy I was to discover that I was right.

First thing I did was connect my Droid 2 Global to my home network via WiFi in order to sniff some of the traffic going to and from the mobile app. This was accomplished by mounting a CIFS drive from my Windows 7 desktop to my router running Tomato based firmware. The share had a binary for TCPDump which I then used to sniff for traffic originating or going to my phone's IP address. After setting this up and performing all of the basic operations I would need in order to "reverse engineer" the data source I got to work on filtering the important bits.

In WireShark I immediately discovered that they had a sub-domain they were using for these operations. All of the web requests that weren't images or for customer metrics and tracking went to this host:

http://www.ows.newegg.com/

Because this API is structured more or less the same as navigating their site and the identifiers are different I decided to start with writing a query builder. Basically the purpose was to allow me to browse to the particular category I was interested in analyzing and filter it down to just a few simple requirements to simplify the analysis.

The first major entry point in the process of browsing to what you're interested in pulling is:

http://www.ows.newegg.com/Stores.egg/Menus

This takes no parameters and provides the main menu:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

[
{
"StoreDepa": "ComputerHardware",
"StoreID": 1,
"ShowSeeAllDeals": true,
"Title": "Computer Hardware"
},
{
"StoreDepa": "PCNotebook",
"StoreID": 3,
"ShowSeeAllDeals": true,
"Title": "PCs & Laptops"
},
{
"StoreDepa": "Electronics",
"StoreID": 10,
"ShowSeeAllDeals": true,
"Title": "Electronics"
},
...

Once you've selected a store to browse the next uri is:

http://www.ows.newegg.com/Stores.egg/Categories/{StoreID}

The only parameter it takes is StoreID which you'll find in the first query. This will return all of the categories within a store. I haven't really explored this very much as I'm only really interested in browsing system memory and SSD's. Using the Computer Hardware store the output is as follows:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

[
{
"Description": "Backup Devices & Media",
"StoreID": 1,
"NodeId": 6642,
"ShowSeeAllDeals": true,
"CategoryType": 0,
"CategoryID": 2
},
{
"Description": "Barebone / Mini Computers",
"StoreID": 1,
"NodeId": 6668,
"ShowSeeAllDeals": true,
"CategoryType": 0,
"CategoryID": 3
},
{
"Description": "CD / DVD Burners & Media",
"StoreID": 1,
"NodeId": 6646,
"ShowSeeAllDeals": true,
"CategoryType": 0,
"CategoryID": 10
},
...

StoreID is included from the parameters of the request. I'm not exactly sure how to describe the purpose of NodeID but it appears to be a distinguishing feature of a category or subcategory. CategoryID is used for filtering results down to a specific category and can be either a root category or a subcategory. CategoryType determines whether CategoryID is a root category or if it contains subcategories. A value of 1 for CategoryType indicates that it is the root category.

Now depending on CategoryType you either move straight to the search query or onto a navigation query. The navigation query is used if there are subcategories:

http://www.ows.newegg.com/Stores.egg/Navigation/{StoreID}/{CategoryID}/{NodeID}

This query takes StoreID, CategoryID and NodeID, which you can get from the category listing of a particular store. It will return a subcategory list. Below is the subcategory listing for the memory category.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

[
{
"Description": "Desktop Memory",
"StoreID": 1,
"NodeId": 7611,
"ShowSeeAllDeals": false,
"CategoryType": 1,
"CategoryID": 147
},
{
"Description": "Flash Memory",
"StoreID": 1,
"NodeId": 8038,
"ShowSeeAllDeals": false,
"CategoryType": 1,
"CategoryID": 68
},
{
"Description": "Laptop Memory",
"StoreID": 1,
"NodeId": 7609,
"ShowSeeAllDeals": false,
"CategoryType": 1,
"CategoryID": 381
},
...

From here you will go to the search query^[3]. At this point it does get a little tricky as the parameters for the query are no longer sent via GET they are instead sent using POST^[4] which basically will require a programmatic method for making a search query. The search query given a category, store and node will list quite a lot of things. The first thing in the list is search filtering parameters, these will allow you to limit the products shown in the listing.

Data being posted is necessary to receive a non-404 response from the server, if you really wanted to you could just send an empty dictionary as this would just query newegg's entire product list. Any of the query options can be omitted, integer values may be omitted by substituting their value with -1.

The parameters you should concern yourself with are as follows along with the URL the data should be posted in JSON format to:

http://www.ows.newegg.com/Search.egg/Advanced

1
2
3
4
5
6
7
8
9

data = {
"SubCategoryId": 147,
"NValue": "",
"StoreDepaId": 1,
"NodeId": 7611,
"BrandId": -1,
"PageNumber": 1,
"CategoryId": 17
}

NValue is a space separated list of NValues from the search parameters. Mind you, you cannot filter against more than one item in any category of search filters. For example in system memory you can't select DDR3 1333 (PC3 10600), DDR3 1333 (PC3 10660) and DDR3 1333 (PC3 10666). The query will return an unsucessful search result. The rest of the parameters are fairly self-explanatory.

The result returned will contain the following elements: RelatedLinkList, CoremetricsInfo, NavigationContentList, PaginationInfo, ProductListItems. CoremetricsInfo and RelatedLinkList can usually be ignored, the elements we're interested in are the NavigationContentList which is a list of search parameters//filters you can apply to the search. PaginationInfo describes how many elements were returned, what page we're on and how many elements there are per page. Last but not least the ProductListItems which provides a list of the products returned by the query along with some basic listing info for each one.

Below is a portion of the NavigationContentList:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

{
"NavigationContentList": [
{
"NavigationItemList": [
{
"SubCategoryId": -1,
"Description": "Free Shipping",
"StoreDepaId": 94,
"NValue": "100007611 600006050 600052012 4808",
"BrandId": -1,
"StoreType": 4,
"ItemCount": 194,
"CategoryId": -1,
"ElementValue": "4808"
},
{
"SubCategoryId": -1,
"Description": "Top Sellers",
"StoreDepaId": -1,
"NValue": "100007611 600006050 600052012 4802",
"BrandId": -1,
"StoreType": -1,
"ItemCount": 39,
"CategoryId": -1,
"ElementValue": "4802"
},
...

This section will also contain a group name:

1
2
3
4
5
6
7
8
9
10
11
12
13

...
"TitleItem": {
"SubCategoryId": -1,
"Description": "Useful Links",
"StoreDepaId": -1,
"NValue": "4800",
"BrandId": -1,
"StoreType": -2,
"ItemCount": 0,
"CategoryId": -1,
"ElementValue": "4800"
}
...

The PaginationInfo and ProductListItem elements will look like the following:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38

...
"PaginationInfo": {
"TotalCount": 233,
"PageNumber": 1,
"PageSize": 20
},
"ProductListItems": [
{
"SellerId": null,
"ItemOwnerType": 0,
"Title": "Crucial Ballistix 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 2133 (PC3 17000) Desktop Memory with Thermal Sensor Model BL2KIT25664FN2139",
"ItemGroupID": 0,
"ReviewSummary": {
"Rating": 5,
"TotalReviews": "[1]"
},
"IsCellPhoneItem": false,
"Discount": null,
"FinalPrice": "$104.99",
"ItemNumber": "20-148-372",
"MappingFinalPrice": "$104.99",
"FreeShippingFlag": true,
"OriginalPrice": "$104.99",
"IsComboBundle": false,
"MailInRebateText": null,
"ProductStockType": 0,
"Model": "BL2KIT25664FN2139",
"ShowOriginalPrice": false,
"Image": {
"FullPath": "http://images17.newegg.com/is/image/newegg/20-148-372-TS?$S125W$",
"SmallImagePath": null,
"ThumbnailImagePath": null,
"Title": null
},
"SellerName": null,
"ParentItem": null
},
...

At this point you might be wondering what good will all this do me if I can't get specifications on an item? Well, you can and here's how: In each ProductListItems element you'll find an ItemNumber, this is essentially the primary key that each product is related to within this interface to newegg's product list. Using the following url you can obtain the full details page on any given item using it's ItemNumber:

http://www.ows.newegg.com/Products.egg/{ItemNumber}/Specification

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77

{
"SpecificationGroupList": [
{
"GroupName": "Model",
"SpecificationPairList": [
{
"Value": "Crucial",
"Key": "Brand"
},
{
"Value": "Ballistix",
"Key": "Series"
},
{
"Value": "BL2KIT25664FN2139",
"Key": "Model"
},
{
"Value": "240-Pin DDR3 SDRAM",
"Key": "Type"
}
]
},
{
"GroupName": "Tech Spec",
"SpecificationPairList": [
{
"Value": "4GB (2 x 2GB)",
"Key": "Capacity"
},
{
"Value": "DDR3 2133 (PC3 17000)",
"Key": "Speed"
},
{
"Value": "9",
"Key": "Cas Latency"
},
{
"Value": "9-10-9-24",
"Key": "Timing"
},
{
"Value": "1.65V",
"Key": "Voltage"
},
{
"Value": "No",
"Key": "ECC"
},
{
"Value": "Unbuffered",
"Key": "Buffered/Registered"
},
{
"Value": "Dual Channel Kit",
"Key": "Multi-channel Kit"
}
]
},
{
"GroupName": "Manufacturer Warranty",
"SpecificationPairList": [
{
"Value": "Lifetime limited",
"Key": "Parts"
},
{
"Value": "Lifetime limited",
"Key": "Labor"
}
]
}
],
"NeweggItemNumber": "N82E16820148372",
"Title": "Crucial Ballistix 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 2133 (PC3 17000) Desktop Memory with Thermal Sensor Model BL2KIT25664FN2139"
}

From this point on you can grab all of the features and specifications of any particular item you're interested in. In the near future I'll be writing a new post for both my memory and SSD analysis scripts using this interface.

The full code for my query builder is as follows, though you should note this was a quick script and is in no way complete or fully functional. As soon as it was to a useable point I moved onto the main point of this whole ordeal. You should also note that this requires CherryPy^[5] and lxml^[6]. The end result of this program is a query which you can use to retrieve a list of products matching the options you've selected. This is mainly to simplify product list selection and to minimalize the need to hardcode in certain values as newegg as a tendency to change things around on a regular basis.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160

import cherrypy, json, urllib, urllib2
from lxml import etree
from lxml.builder import E

class QueryBuilder(object):
def index(self):
request = urllib2.urlopen("http://www.ows.newegg.com/Stores.egg/Menus")
response = request.read()
data = json.loads(response)

body = E.body()

ul = E.ul()
for store in data:
ul.append(E.li(E.a(
store['Title'],
href= '/Store?StoreID={}'.format(store['StoreID'])
)))

page = E.html(E.body(ul))

return etree.tostring(page, pretty_print=True)
index.exposed = True

def Store(self, StoreID=None):
if StoreID is not None:
request = urllib2.urlopen("http://www.ows.newegg.com/Stores.egg/Categories/{}".format(StoreID))
response = request.read()
data = json.loads(response)

body = E.body()

ul = E.ul()
for category in data:
if category['CategoryType'] == 1:
ul.append(E.li(E.a(
category['Description'],
href='/Search?StoreID={}&CategoryID={}&NodeID={}'.format(StoreID, category['CategoryID'], category['NodeId'])
)))
else:
ul.append(E.li(E.a(
category['Description'],
href='/Category?StoreID={}&CategoryID={}&NodeID={}'.format(StoreID, category['CategoryID'], category['NodeId'])
)))

page = E.html(E.body(ul))

return etree.tostring(page, pretty_print=True)
else:
return "Invalid parameters."
Store.exposed = True

def Category(self, StoreID, CategoryID, NodeID):
if None not in [StoreID, CategoryID, NodeID]:
request = urllib2.urlopen("http://www.ows.newegg.com/Stores.egg/Navigation/{}/{}/{}".format(StoreID, CategoryID, NodeID))
response = request.read()
data = json.loads(response)

body = E.body()

ul = E.ul()
for subcategory in data:
ul.append(E.li(E.a(
subcategory['Description'],
href= '/Search?StoreID={}&CategoryID={}&SubCategoryID={}&NodeID={}'.format(StoreID, CategoryID, subcategory['CategoryID'], subcategory['NodeId'])
)))

page = E.html(E.body(ul))

return etree.tostring(page, pretty_print=True)
else:
return "Invalid parameters."
Category.exposed = True

def Search(self, StoreID=None, CategoryID=None, SubCategoryID=None, NodeID=None):
url = "http://www.ows.newegg.com/Search.egg/Advanced"
data = {
"IsUPCCodeSearch": False,
"IsSubCategorySearch": True,
"isGuideAdvanceSearch": False,
"StoreDepaId": StoreID,
"CategoryId": CategoryID,
"SubCategoryId": SubCategoryID,
"NodeId": NodeID,
"BrandId": -1,
"NValue": "",
"Keyword": "",
"Sort": "FEATURED",
"PageNumber": 1
}

params = json.dumps(data).replace("null", "-1")
request = urllib2.Request(url, params)
response = urllib2.urlopen(request)
data = json.loads(response.read())

if data['NavigationContentList'] is None:
return etree.tostring(E.pre(json.dumps(data, indent=4)), pretty_print=True)

body = E.body()

form = E.form(name='PowerSearch', action='GenerateURL', method='GET')

table = E.table()
form.append(table)
for section in data['NavigationContentList']:
index = 0
tr = E.tr(E.td(section['TitleItem']['Description'], colspan='3'))
table.append(tr)
for option in section['NavigationItemList']:
if index % 3 == 0:
tr = E.tr()
table.append(tr)
index += 1
checkbox = E.td(E.input(option["Description"], type="checkbox", name=section['TitleItem']['Description'].replace(" ", ""), value=option['NValue']))
tr.append(checkbox)

for param, value in [('StoreID', StoreID), ('CategoryID', CategoryID), ('SubCategoryID', SubCategoryID), ('NodeID',NodeID)]:
try:
form.append(E.input(type='hidden', name=param, value=value))
except KeyError:
pass
form.append(E.input(type='submit', value='Submit'))
page = E.html(E.body(form))

return etree.tostring(page, pretty_print=True)
Search.exposed = True

def GenerateURL(self, StoreID=None, CategoryID=None, SubCategoryID=None, NodeID=None, **kwargs):
NValue = set([])
for arg in kwargs:
if type(kwargs[arg]) == list:
for value in kwargs[arg]:
NValue.add(value)
else:
NValue.add(kwargs[arg])

NValue = list(NValue)
NValue.sort()
if StoreID is None:
StoreID = -1
if CategoryID is None:
CategoryID = -1
if SubCategoryID is None:
SubCategoryID = -1
if NodeID is None:
NodeID = -1
data = {
"StoreDepaId": int(StoreID),
"CategoryId": int(CategoryID),
"SubCategoryId": int(SubCategoryID),
"NodeId": int(NodeID),
"BrandId": -1,
"NValue": ' '.join(NValue),
"PageNumber": 1
}
return etree.tostring(E.pre(json.dumps(data, indent=4)), pretty_print=True)
GenerateURL.exposed = True

cherrypy.quickstart(QueryBuilder())

And iOS devices I assume as well.
Because lets face it, that would be stupid.
... or get to the search query from selecting a root category in the main category listing for a store
At least this is the method used by the mobile app.
CherryPy: CherryPy is a pythonic, object-oriented HTTP framework.
lxml: A Pythonic binding for the C libraries libxml2 and libxslt.

Choosing an SSD (A more different S)

bemasher — Mon, 13 Dec 2010 06:14:14 +0000

I've been periodically going back and revisiting the results for my SSD analysis script for newegg.com. The last few times I ran it I noticed that it was broken. It looks like newegg has modified a few things in their power search results page. One thing which is a little obnoxious^[1] is that they no longer include the capacity in the description of the item or as a feature in the feature list when viewing the results page. This only seems to be an issue on the SSD page although I can't figure out why they decided it didn't need to be there in the first place. I see it this way: SSD's are first and foremost a storage device, you'd think that one of the most important features that should be listed with every SSD is the capacity at least.

Anyway, this change broke my script which I had been meaning to rewrite since regular expressions are definitely not the most efficient or cleanest way to parse HTML. I've been working with XML a more often lately despite my original prejudice against it for being a really bloated way to transfer data. One thing I discovered that makes XML a lot less painful is XPath^[2] which is an incredibly useful "language" for selecting data from an XML document.

Once I had gone through and read several tutorials and references about XPath I set out to use it in writing a show calendar script which parses data from tvrage.com's XML API. After that useful exercise I realized I could very easily and cleanly apply it to my SSD analysis script. Since HTML is similar in nature to XML^[3] I set out to parse Newegg's results page using XPath. This presented the first problem: Newegg's page isn't strictly XML or even XHTML for that matter. After a great deal of googling and research I landed on the lxml^[4] website which as it turns out has an HTML parser for navigating and extracting data from HTML in the same way you would from an xml.etree.ElementTree^[5]. With this in mind I immediately began rewriting the script.

First off lets consider my criteria for a "good" SSD on Newegg. The SSD can be either the typical 2.5" form factor, or a PCI-Express card^[6]. The interface can be SATAII, SATAIII or PCI-Express. Capacity must be greater than or equal to 120GB^[7]. Last but not least, the disk should be sub $300^[8].

The above requirements give us the following power search^[9] which we will be using as the source for the script:

1
2
3
4
5
6

url = "http://www.newegg.com/Product/ProductList.aspx?Submit=Property&N=100008" + \
"120&IsNodeId=1&maxPrice=300&OEMMark=1,0&PropertyCodeValue=4213:30854,421" + \
"3:41472,4213:47725,4214:46019,4214:72313,4214:57574,4214:58118,4214:3941" + \
"6,4214:47732,4214:30849,4214:47171,4214:46300,4214:77918,4214:72311,4214" + \
":77919,4214:55178,4214:47733,4214:57755,4214:44038,4215:55552,4215:47726" + \
",4215:41071&bop=And&Pagesize=100"

Now the first thing that made me cringe as I was rewriting this was the fact that I would basically have no choice but to load each individual product page from the results page as capacity is no longer included in either the description or the features list of each product in the results page. Eventually I will get around to multi-threading this to make it a little less painful, or I'll get lucky and Newegg will add the capacity feature back to the item listing in power searches for SSD's. The following is the full source code of the parser:

import re, math
from lxml import etree

url = "http://www.newegg.com/Product/ProductList.aspx?Submit=Property&N=100008" + \
"120&IsNodeId=1&maxPrice=300&OEMMark=1,0&PropertyCodeValue=4213:30854,421" + \
"3:41472,4213:47725,4214:46019,4214:72313,4214:57574,4214:58118,4214:3941" + \
"6,4214:47732,4214:30849,4214:47171,4214:46300,4214:77918,4214:72311,4214" + \
":77919,4214:55178,4214:47733,4214:57755,4214:44038,4215:55552,4215:47726" + \
",4215:41071&bop=And&Pagesize=100"

featureMap = {
'Capacity': 'capacity',
'Sequential Access - Write:': 'write',
'Sequential Access - Write': 'write',
'Sequential Access - Read:': 'read',
'Sequential Access - Read': 'read',
'Interface Type': 'interface',
'Brand': 'brand',
'Model': 'model',
'Series': 'series'
}

speed_re = re.compile(r'(\d+)\s?MB/s')
capacity_re = re.compile(r'(\d+)GB')

parser = etree.HTMLParser()
# tree = etree.parse("temp.html", parser)
tree = etree.parse(url, etree.HTMLParser())
root = tree.getroot()

items = []

for node in root.findall(".//div[@class='itemCell']"):
item = {}

# Get link
link = node.find(".//a[@title='View Details']")
item["link"] = link.attrib["href"]

# Get feature list (loads each item's url, should multi-thread this in the future)
itemPage = etree.parse(item["link"], etree.HTMLParser()).getroot()
featureList = map(lambda n: n.text, itemPage.findall(".//fieldset/dl/dt"))
valueList = map(lambda n: n.text, itemPage.findall(".//fieldset/dl/dd"))
features = zip(featureList, valueList)
for feature, value in features:
if value is not None and feature in featureMap:
# If it's a speed feature parse out the speed
if featureMap[feature] in ("read", "write"):
item[featureMap[feature]] = min(map(lambda x: int(x), speed_re.findall(value)))
# If it's a capacity feature, parse out the capacity
elif featureMap[feature] == "capacity":
item[featureMap[feature]] = min(map(lambda x: int(x), capacity_re.findall(value)))
# If the value doesn't need to be parsed, just store the value in item
else:
item[featureMap[feature]] = value.strip()

# Get price
price = map(lambda n: n.text, node.findall(".//li[@class='priceFinal']/*"))
item["price"] = float(''.join(price[1:]))

# Only add the item if it has the features we need in it
if "read" in item and "write" in item and "capacity" in item and "series" in item:
score = (item["read"] * item["write"] * item["capacity"]) / ((math.log(abs(item["read"] - item["write"])) + 1) + item["price"])
item["score"] = score
items.append(item)

sorted = {}
for item in items:
# Open addressing like in a hash table, so we don't wind
# up with any collisions, unlikely but good practice anyway
score = item["score"]
while score in sorted:
score += 1
sorted[score] = item

sortOrder = sorted.keys()
sortOrder.sort()
sortOrder.reverse()

headers = ['brand', 'series', 'model', 'link', 'interface', 'price', 'capacity', 'read', 'write', 'score']
print '\t'.join(headers)
for key in sortOrder:
item = sorted[key]
print '\t'.join(map(lambda x: str(item[x]), headers))

At this point if you've gone through and read the entire script you'll probably notice that I've made a slight change to the scoring equation, it has been changed from the following:

To the following:

I discovered that using the difference in read//write speed heavily penalized drives with anything greater than 10MB/s difference. So I figured that it may be a little more subtle to simply penalize drives based on the magnitude of the difference.

Now you're probably wondering: "When is this blathering idiot going to get to the damned results already?". And you'd be pleasantly surprised to know that I'm getting to them as you waste your time reading this.

Manufacturer:	OCZ	OCZ	G.Skill	OCZ
Series:	RevoDrive	Vertex 2	Phoenix Pro Series	Agility 2
Capacity:	120GB	180GB	120GB	120GB
Read:	540MB/s	285MB/s	285MB/s	285MB/s
Write:	490MB/s	275MB/s	275MB/s	275MB/s
Item:	N82E16820227578^[10]	N82E16820227602^[11]	N82E16820231378^[12]	N82E16820227593^[13]
Price:	$299.99	$294.99	$214.99	$214.99

As you can see the RevoDrive far out-scores all the rest of the SSD's considered in this analysis. The main reason is that they've essentially included two 60GB SSD's on the same card and you're expected to perform software raid on them in your own system^[14]. Despite the incredible speeds they boast I don't think I would purchase one of these to use as my OS//Program disk because compatibility is a major limitation. You must be sure that your motherboard's BIOS supports booting via PCI-Express cards. And last but not least, the main reason I would pass up this card is the lack of TRIM support. As far as I can tell these cards do not support TRIM which is a major downside as far as I'm concerned.

The second disk in the list is the OCZ Vertex 2 180GB version. I'd probably skip this one just because I don't really consider the extra 60GB worth the extra $80.

Which leaves me with the last two disks which are as far as my analysis is concerned, identical. If you take into account the detailed features you'll notice that the G.Skill claims 50k IOPS on the 4k Random write test which seems a bit... optimistic. The OCZ makes no such claim and as far as I'm concerned both disks are more less the same thing. So it's pretty much up to brand preference at this point.

I've already sent feedback to them suggesting that they fix this.
Only if the XML parser you're using supports it, which it seems is not a whole lot of them. At least not all of them support the full specification which is annoying since nobody really seems to document which bits and pieces they support and which whey don't.
Although not necessarily XML depending on the particular doctype you've chosen, Newegg's is transitional HTML.
lxml: http://codespeak.net/lxml/
xml.etree.ElementTree: http://docs.python.org/library/xml.etree.elementtree.html
Some of the PCI-Express SSD's are stupidly fast and more expensive except that it doesn't look like any of them support TRIM yet which is a major problem for me.
It is rare that I have a matured (read: haven't reformatted in a while) install of windows along with all of my most commonly used programs and games that exceeds 60GB so I estimate that doubling this should accommodate for any sudden urges to install really big things.
I can't really justify spending much more than $300 on a single storage device. It had better be one hell of a storage device if I ever find myself spending more than $300 on it.
This will likely need to be updated at least once a month as Newegg is constantly adding new criteria and changing things.
OCZ RevoDrive
OCZ Vertex 2
G.SKILL Phoenix Pro Series
OCZ Agility 2
They show up as two separate physical devices despite being located on the same card.

Automagic TV Show Calendar

bemasher — Sat, 04 Dec 2010 13:11:57 +0000

A little while ago I was browsing the web and discovered a website called tvrage.com^[1] which seems to be the definitive online TV guide. I didn't originally enter the site on the main index but on a page describing the functionality of an XML API^[2] they host for accessing their database of TV shows.

To me, this is like opening presents on christmas day. Just imagine the possibilities! I immediately began exploring the kind of data they provide. The very first idea I had was to use this to create events on my google calendar automatically for unaired episodes of my favorite TV shows.

I've previously written python scripts that interface with gdata but I find their implementation for python to be kind of cumbersome to deal with so I began researching their Protocol API^[3]. At first I wasted a lot of time attempting to build the necessary XML structures to add events and the like. This got old very fast and I decided to just give JSON-C^[4] a try. Turns out you can use the built-in JSON module in python for creating the necessary structures.

For parsing the results I got from tvrage I ended up using python's xml.etree.ElementTree which was simple enough to setup to retrieve only the information for each episode I was interested in.^[5]

I had a bit of trouble initially with adding events to google calendar. This stemmed from the fact that google often will return an HTTP Redirect which includes a url with an appended gsession attribute which you're supposed to resubmit the exact data from the first request to. Once I figured this out it was turtles all the way down. I even managed to get the whole script multi-threaded to speed things up since it's impossible to perform batch-requests with JSON-C.

I should note that for the configuration file the calendar should be the "Calendar ID" for the calendar that can be found by looking at the settings page for the individual calendar, it is grouped with the XML and iCal feeds.

ShowList.txt:^[6]

1
2
3
4
5
6
7
8
9
10
11
12

Castle 19267
House 3908
Bones 2870
Big Bang Theory, The 8511
Mentalist, The 18967
Rizzoli & Isles 24996
Venture Bros., The 6270
Top Gear 6753
Mythbusters 4605
Archer 23354
NCIS 4628
Community 22589

Config.cfg:

1
2
3
4

[Credentials]
username = someuser@gmail.com
password = somebase64encodedpassword
calendar = somecalendarid@group.calendar.google.com

AirDate.py:

import urllib2, urllib, json, ConfigParser, base64
from datetime import date
from xml.etree import ElementTree
from threading import Thread

calendar = ""
header = {}

# Thread for retrieving a list of episodes for a given show_id
class airDate(Thread):
# Initialize thread and set some local attributes
def __init__(self, show_name, show_id):
Thread.__init__(self)
self.show_name = show_name
self.show_id = show_id

# Get episode list from tvrage.com based on the show_id
def run(self):
# Retrieve XML episode_list from tvrage.com
xml_data = urllib2.urlopen("http://services.tvrage.com/feeds/episode_list.php?sid=%s" % self.show_id).read()
# Pares XML into ElementTree.Element()
xml_tree = ElementTree.fromstring(xml_data)
self.result = []

# For each season
for season in xml_tree.findall("Episodelist/Season"):
# Get the season number
season_num = int(season.get("no"))
# For each episode in the episode list
for episode in season.findall("episode"):
# Get episode number and title
episode_num = int(episode.find("seasonnum").text)
episode_title = episode.find("title").text

# Build the episode code S##E##
episode_code = "S%02dE%02d" % (season_num, episode_num)

# Parse the airdate into year, month and day
year, month, day = map(lambda x: int(x), episode.find("airdate").text.split("-"))
try:
episode_airdate = date(year, month, day)
today = date.today()
# If episode hasn't aired yet
if episode_airdate >= today:
# Add episode to results list
self.result.append("%s %s - %s" % (str(episode_airdate), self.show_name, episode_code))
except ValueError:
# If the airdate is invalid (tvrage.com sometimes
# includes 00's for unknown sections of the date
pass

class addEvent(Thread):
# Thread for adding events to google calendar

# Initialize thread and set local episode variable
def __init__(self, episode):
Thread.__init__(self)
self.episode = episode

# Add new entry to google calendar
def run(self):
# Build entry structure
entry = {"data": {"details": self.episode, "quickAdd": True}}
# Convert to JSON
entry = json.dumps(entry)

# Build request including necessary headers and data
calReq = urllib2.Request("http://www.google.com/calendar/feeds/%s/private/full?alt=jsonc" % (calendar), entry, header)
# Execute the request
calRes = urllib2.urlopen(calReq)
# Get the redirect url (gsession appended)
redirectReq = urllib2.Request(calRes.geturl(), entry, header)
try:
redirectRes = urllib2.urlopen(redirectReq)
except HTTPError:
# If we get some sort of HTTP error code
# skip entry, can always run again
pass

# Get list of events already added to
# the calendar from previous executions
def getExistingEpisodes(header):
# Get JSON-C representation of calendar
calReq = urllib2.Request(url="https://www.google.com/calendar/feeds/%s/private/full?alt=jsonc" % (calendar), headers=header)
calRes = urllib2.urlopen(calReq)

# Parse JSON-C
data = json.loads(calRes.read())
# If the calendar has events on it
if "items" in data["data"]:
# Get the list of events
events = data["data"]["items"]
existing_episodes = []
# For each event
for event in events:
# Append just the title of the event to the results
existing_episodes.append(event["title"])

return existing_episodes
else:
# We don't have any events on this calendar
# so just return an empty list
return []

if __name__ == '__main__':
# Open the configuration file and get the necessary
# credentials and settings
config = ConfigParser.ConfigParser()
config.readfp(open("Config.cfg"))
username = config.get("Credentials", "username")
password = config.get("Credentials", "password")
# Password is stored as base64 encoded string just so
# we don't have our password sitting out in plain sight
password = base64.b64decode(password)
calendar = config.get("Credentials", "calendar")

# Build loginData structure, this is used to get
# authentication data from google
loginData = {
"Email": username,
"Passwd": password,
"source": "BeMasher-ETR-2",
"service": "cl"
}

# Encode the loginData for usage in a url
loginData = urllib.urlencode(loginData)
# Get authentication data
gdataLogin = urllib2.urlopen("https://www.google.com/accounts/ClientLogin", data=loginData)
SID, LSID, Auth = gdataLogin.read().splitlines()

# Build header structure, this will be used for
# all requests to google calendar from now on
header = {
"Authorization": "GoogleLogin %s" % (Auth),
"GData-Version": 2,
"Content-Type": "application/json"
}

# Open a list of the shows we're interested in
# Stored as "show_name\tshow_id", one per line
show_list = open("ShowList.txt")
jobs = []
for line in show_list:
show = line.strip().split("\t")
jobs.append(show)

# Get a list of existing events from previous
# executions so we don't wind up with duplicates
existingEpisodes = getExistingEpisodes(header)

threadQueue = []
# For each episode we've retrieved that is unaired
for job in jobs:
show_name, show_id = job
# Create an instance of the airDate thread
thread = airDate(show_name, show_id)
# Start it
thread.start()
# Add it to the threadQueue
threadQueue.append(thread)

episodes = []
# While we've still got running threads
while len(threadQueue) > 0:
# Get a thread from the queue
thread = threadQueue.pop()
# Block until it completes
thread.join()
# For each episode in the results
for episode in thread.result:
# If it hasn't already been added to google calendar
if episode[11:] not in existingEpisodes:
print episode
# Add to list of episodes that need events created
episodes.append(episode)

# For each episode that doesn't have an
# event on google calendar already
for episode in episodes:
# Create an addEvent thread, start it
# and add it to the threadQueue
thread = addEvent(episode)
thread.start()
threadQueue.append(thread)

# While we still have threads running
while len(threadQueue) > 0:
# Get a thread from the queue
thread = threadQueue.pop()
# Block until it completes
thread.join()

This was all done shortly before I discovered that tvrage.com also provides iCal feeds for your favorite shows provided that you register and add some to your list. Unfortunately the iCal feed they generate creates events for exact air times of each episode which I'm not really all that concerned about. So I use this script still to add all-day events for each episode which is easier to view//see when there's a new episode.

I did write another script using their XML API but that will have to wait for another post.

http://tvrage.com/
http://services.tvrage.com/
Data API Developer's Guide: The Protocol
Google's own flavor of JSON which is almost identical to plain old JSON.
I only really needed the original air date, title, season number and episode number.
You can find the show_id via the show search found on their XML API page.

Choosing an SSD (Update)

bemasher — Sun, 10 Oct 2010 02:42:58 +0000

My brother is in the planning stages of building a new desktop. One of the things he's planning on doing differently from his last build^[1] is using an SSD for OS + Programs.

I had mentioned to him previously that I a wrote a program for helping to choose an SSD based on what SSD's are meant for and are good at doing. So he asked if I could recommend him one. Below are the results of the latest run of my script based on the most current listings^[2] of SSD's newegg offers.

Manufacturer:	OCZ	G.Skill	OCZ
Series:	Agility 2	Phoenix Pro	Vertex 2
Capacity:	120GB	120GB	120GB
Read:	285MB/s	285MB/s	285MB/s
Write:	275MB/s	275MB/s	275MB/s
Item:	N82E16820227543^[3]	N82E16820231378^[4]	N82E16820227551^[5]
Price:	$235.99	$239.99	$240.00

It looks like OCZ has two of the top three places this run and G.Skill is still maintaining one of the top three from before. Between the 3 of them I think i would likely still go for the G.Skill just because of personal preference despite there not really being any significant differences between the three. Excepting price of course.

Which incidentally was right when SSD's were just becoming available to the average consumer.
As of this date 10/09/2010.
OCZ Agility 2
G.Skill Phoenix Pro
OCZ Vertex 2

Choosing an SSD

bemasher — Thu, 12 Aug 2010 06:02:27 +0000

Before I started my new job I had an inordinate amount of free time and for a majority of that time, nothing to spend it doing^[1]. I was still thinking about my desktop wishlist^[2] and about choosing a better SSD than the one I had previously selected^[3].

A long time ago when I was following the HDD market since I was looking to buy some bulk storage I wrote a php script which loaded newegg's product list based on some search parameters you provided newegg's productlist.xml^[4]. The script would then parse the list and produce a list sorted based on price per gigabyte. Which is useful when you're in the market for capacity^[5].

I decided to do more or less the same thing with SSD's except this time I did it in python since I'm rusty on PHP and I didn't want to mess with setting up a web server to test on. So I got started by doing a power search on newegg for the specific flavor of SSD I was looking for.

The search parameters are as follows:

2.5" Form Factor
SATA II/III
120GB or Greater
Less than $300
Retail or OEM
Support TRIM Command

As of this writing those particular search parameters narrows the result to 17 SSD's. Now comes the code. Before I started coding I needed some way to sort them according to what I thought was important. The metric is as follows:

$$\frac{\text{Read} \times \text{Write} \times \text{Capacity}}{|\text{Read} - \text{Write}| \times \text{Price}}$$

After looking closer at the scores this produces I noticed that it heavily penalizes drives with huge differences between read and write speeds which effectively weeds out drives that still have acceptable read//write speeds. So I removed that section of the metric producing:

The basic idea behind this scoring measure is that sequential read and write speeds are important, as well as capacity. Price and difference between sequential read//write are considered bad^[6]. In the equation read and write refer to sequential read and write speeds. The ratio of these will produce a score of the SSD's overall performance for capacity, read//write speeds and price.

The code is relatively simple in purpose. Load the data and parse it into a dictionary then sort based on the metric above.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59

import urllib2, re

# url = "
# http://www.newegg.com/Product/ProductList.aspx?Submit=Property&Subcatego
# ry=636&Description=&Type=&N=100008120&IsNodeId=1&srchInDesc=&MinPrice=&M
# axPrice=&OEMMark=1&OEMMark=0&PropertyCodeValue=4213:30854&PropertyCodeVa
# lue=4214:30848&PropertyCodeValue=4214:39416&PropertyCodeValue=4214:30849
# &PropertyCodeValue=4214:39415&PropertyCodeValue=4215:55552&PropertyCodeV
# alue=4215:41071&PropertyCodeValue=4215:46319"

# data = open("temp.html", "w")
# data.write(urllib2.urlopen(url).read())
# data.close()
raw = open("temp.html").read()

item_re = re.compile(r'

(.*?)
')
feature_re = re.compile(r"

(.*?)

")
feature_list_re = re.compile(r'(.*?)\s?\#?\s?:\s?\s?(.*?)')
speed_re = re.compile(r"(up to )?(\d+).*?MB/s")
capacity_re = re.compile(r"(\d+)GB")
price_re = re.compile(r"\$(\d+)^.(\d+)")

item_list = []
valid = ['Read', 'Item', 'Interface', 'Capacity', 'Model', 'Write', 'Size']

for item in item_re.findall(raw):
current = {}
no_label = []
features = feature_re.findall(item)
current["Size"] = features[0]
current["Capacity"] = features[1]
current["Interface"] = features[2]

for feature in feature_list_re.findall(item):
if feature[1].find("\r") != -1:
current[feature[0]] = feature[1].split("\r")[0]
else:
current[feature[0]] = feature[1]
current["Read"] = int(speed_re.findall(current["Sequential Access - Read"])[0][1])
current["Write"] = int(speed_re.findall(current["Sequential Access - Write"])[0][1])
current["Capacity"] = int(capacity_re.findall(current["Capacity"])[0])
for feature in current.keys():
if feature not in valid:
del current[feature]
current["Price"] = float('.'.join(price_re.findall(item)[0]))
current["Item"] = "http://www.newegg.com/Product/Product.aspx?Item=%s" % (current["Item"])
item_list.append(current)

sorted = {}
for item in item_list:
ratio = (item["Read"] * item["Write"] * item["Capacity"]) / (item["Price"])
sorted[ratio] = item

sort_order = sorted.keys()
sort_order.sort()
sort_order.reverse()
for key in sort_order:
#print '\t'.join(map(lambda x: str(x), sorted[key].keys()))
print '\t'.join(map(lambda x: str(x), sorted[key].values()))

Now given that there is quite a lot of data to present and analyze all at once I've decided it would be easiest to just provide you with a pretty graph^[7]:

If you look closely at the scores of all the disks in the query, you'll notice that this is a noticeable gap between the top 3 and the rest. They are as follows:

Manufacturer:	A-DATA	Patriot	G.Skill
Series:	S599	Inferno	Phoenix Series
Capacity:	128GB	120GB	120GB
Read:	280MB/s	285MB/s	285MB/s
Write:	270MB/s	275MB/s	275MB/s
Item:	N82E16820211471^[8]	N82E16820220510^[9]	N82E16820231372^[10]
Price:	$295.99	$289.99	$299.00

I noticed that if you ignore capacity in the metric then the Patriot Inferno is the clear winner here. So as it turns out the Western Digital SiliconEdge I had selected when I first wrote the wishlist wasn't the best drive for my needs. But then I've always had a soft-spot for Western Digital. But now I'm convinced that the Patriot Inferno is the SSD I'll be getting unless by the time I get around to buying one there are better options^[11].

Nothing worth-while anyway
See previous post: Wishlist.
Western Digital SiliconEdge 128GB SSD
Which no longer exists in it's original form.
Which I was.
Although we're excluding read//write speed difference.
Scores have been normalized to 100%.
A-Data S599
Patriot Inferno
G.Skill Phoenix Series
Which there probably will be.

Matplotlib and Live Data: A Tale of Two Technologies

bemasher — Thu, 29 Jul 2010 04:46:53 +0000

Being unemployed over the summer is never usually a good thing for me. I get bored very easily if I don't have something to occupy myself with. This last bout of boredom led me to unpack some of my electronics. Dusted off my multimeter, Arduino and a digital thermometer I bought a little while ago. Figured I could use these to solve one of my current problems.

Living in Laramie usually subjects people to harsh winters which leaves most housing developments without central air conditioning installed since, well it's never really needed except maybe one or two days over the summer where it gets above 85 ^oF. This summer has apparently been hotter than previous summers and It's left my condo in an "uncomfortable state". Mind you I'm used to living in hot weather so this isn't such a terrible thing to me, I'm used to it.

What I'm not used to is not having AC and it cooling off enough at night that it's worthwhile to open a few windows and stick a fan in one of them. Which leaves me with this problem: When is the optimal time to open the windows and turn on the fan to get my condo cooled off earliest//fastest?

In comes my Arduino + digital thermometer^[1]. Once I rigged up the proper power//data connections on a breadboard for my Arduino I set out to find code for the thermometer. I've setup the thermometer with a sketch on my Arduino before I just didn't feel like wasting a few hours trying to do it from scratch again. Soon enough I found some code^[2] that worked perfectly. So I trimmed out some code I didn't need for the project and set it up to just write the temperature as fast as possible^[3] to the serial port it's connected to.

After that I wrote a logging program on my desktop in Python to record temperatures sent via serial to my desktop. The program is incredibly simple and uses the pySerial library^[4] to read temperatures from the serial port of my desktop and append them to a temperature log. I used a simple windows command to do this since it wouldn't lock the file so I could read data from it simultaneously. There are still occasionally collisions with the processing program locking the file and the logger not being able to write the data to the file but these are rare enough that it's negligible in my situation.

1
2
3
4
5

import serial, os

ser = serial.Serial(2)
while True:
os.system("echo %s>>out.txt" % (ser.readline().strip()))

The next step in this project was visualizing the data. I've used matplotlib^[5] before and I was thinking this time I would like to see if I could write the program to update data live as it recieves it. My first foray into this goal was a miserable disaster. Most of the solutions I could find involved just setting up an infinite loop with a short time delay in it. Which works great except that it sleeps the thread running the plot which makes it impossible to resize the plot or do anything at all with the GUI for that matter. So obviosly this wouldn't work at all.

After poking around for different solutions to this and crashing my computer once from spawning an infinite number of instances of the plot I gave up for a bit, only to discover that there was an example in the documentation which wasn't obviously named. I quickly discovered the best way to do this. I even added some pretty annotations and such.

import gobject
import matplotlib
matplotlib.use('GTKAgg')

import matplotlib.pyplot as plt

current_pos = 0
temps = []
pad = 5.0

f = plt.figure()

def update(vars):
# Unpack variables that need to be persistent between
# executions of this method.
temps = vars[0]
current_pos = vars[1]
pad = vars[2]

# Open the data file and get any new data points since
# the last time we read from this file
data = open("out.txt", "r")
data.seek(current_pos)
new_temps = map(lambda x:
float(x) * (1 + 4.0/5.0) + 32.0,
data.read().split("\n")[:-1])
current_pos = data.tell()
data.close()

# If we got new data then append it to the list of
# temperatures and trim to 750 points
if len(new_temps) > 0:
temps.extend(new_temps)
temps = temps[-750:]

f.clear()
f.suptitle("Live Temperature")
a = f.add_subplot(111)
a.grid(True)
l, = a.plot(temps)
plt.xlabel("Time (Seconds)")
plt.ylabel(r'Temperature $^{\circ}$F')

# Get the minimum and maximum temperatures these are
# used for annotations and scaling the plot of data
min_t = min(temps)
max_t = max(temps)

# Add annotations for minimum and maximum temperatures
a.annotate(r'Min: %0.2f$^{\circ}$F' % (min_t),
xy=(temps.index(min_t), min_t),
xycoords='data', xytext=(20, -20),
textcoords='offset points',
bbox=dict(boxstyle="round", fc="0.8"),
arrowprops=dict(arrowstyle="->",
shrinkA=0, shrinkB=1,
connectionstyle="angle,angleA=0,angleB=90,rad=10"))

a.annotate(r'Max: %0.2f$^{\circ}$F' % (max_t),
xy=(temps.index(max_t), max_t),
xycoords='data', xytext=(20, 20),
textcoords='offset points',
bbox=dict(boxstyle="round", fc="0.8"),
arrowprops=dict(arrowstyle="->",
shrinkA=0, shrinkB=1,
connectionstyle="angle,angleA=0,angleB=90,rad=10"))

# Set the axis limits to make the data more readable
a.axis([0,len(temps), min_t - pad,max_t + pad])

f.canvas.draw_idle()

# Repack variables that need to be persistent between
# executions of this method
vars = {0: temps, 1: current_pos, 2: pad}

return True

vars = {0: temps, 1: current_pos, 2: pad}

# Execute update method every 500ms
gobject.timeout_add(500, update, vars)

# Display the plot
plt.show()

This code generates a plot which updates every 500ms. This is based on an example in the matplotlib examples^[6]. An example of the program's output can be seen below.

I imagine that I could have made this simpler by not using the GTK libraries which are a pain to install since there are 3 or 4 modules you have to install in order to make all this work including the GTK+ runtime. I may come back later and post a version written using TK since it can be used without installing extra modules and stuff.

DS18S20 Digital Thermometer Datasheet
Temperature Measurement using the Dallas DS18B20 by Peter H. Anderson
Somewhere in the range of 750ms between readings since it is in parasite mode, may change this later to run in non-parasite mode.
pySerial Python Library
matplotlib Python Library
Animation example code: simple_anim_gtk.py

WriteMonkey and Markdown

bemasher — Mon, 28 Jun 2010 03:14:01 +0000

Recently Download Squad had a post^[1] about a practical way to get features and support for open-source programs, specifically through donations. The post was about a program called WriteMonkey which is a minimalistic writing program that the author had originally written about previously^[2]. Think of the best code editing program you know of, mine is Notepad++^[3]. Now take that program and refactor it specifically for writing articles or blog posts, you've just created WriteMonkey^[4].

Something that interested me about WriteMonkey was the Download Squad author's post specifically mentioned writing posts using Markdown^[5] syntax. Markdown is a simple plain-text syntax which is parsed into html removing the need to tediously enter html^[6] as you write. At first glance it didn't really seem like it would really help all that much when it came to writing blog posts. But I was completely wrong and am better off for it. Now the especially useful part is that WriteMonkey supports this completely as well as having a very useful shortcut for parsing and copying html straight from Markdown source. This is incredibly useful since I can then just go to my website and paste the resulting html into a blog post and hit save and be done with it.

As I looked through the program I realized, this is much much more than just a Markdown IDE. It includes all sorts of useful features like a "progress bar" which tells you how far along you are in a certain quota you specify in the preferences. This led me to write a little bit of SQL^[7] to calculate the average word-count of posts in my blog. Excluding the outliers it came out to ~350 words per post. So I just set the quota to 350 words and it displays a bar at the top or bottom of the screen depending on what you choose showing your current progress on the quota.

It also does several other useful things like displaying current battery life as a percentage in the progress bar, showing the file you're writing in. There's also this feature called repository//main. This allows you to store text clippings in repository and then write the blog post in main. When exported as html the repository is ignored and only main is copied. Makes it useful to write notes and such in the middle of authoring a post to keep with everything you write and it's easy enough to switch between the two to make it useful. For this post I just made a list of points I wanted to cover.

After using WriteMonkey for an hour or so I think I've found the new environment I'll be writing all my posts in for the foreseeable future.

Download Squad: Amazing software tip: Pay free software developers to get stuff fixed!
Download Squad: WriteMonkey is an unbelievable full-screen text editor
Notepad++
WriteMonkey
Markdown
HTML: HyperText Markup Language, is the predominant markup language for web pages.
SQL: Structured Query Language

FreeNAS Users Rejoice!

bemasher — Tue, 04 May 2010 06:06:10 +0000

Unetbootin^[1] now supports FreeNAS! Take a look at these awesome little snippets of code:

1
2
3
4
5
6
7
8
9
10
11
12
13
14

//distrolst.cpp
if (nameDistro == "FreeNAS") {
if (isarch64) {
cpuarch = "amd64";
} else {
cpuarch = "i386";
}
instIndvfl("memdisk", QString("%1ubnkern").arg(targetPath));
if (islivecd) {
downloadfile(QString("http://downloads.sourceforge.net/sourceforge/lubi/FreeNAS-%1-LiveCD-%2.img.gz").arg(cpuarch, relname), QString("%1ubninit").arg(targetPath));
} else {
downloadfile(QString("http://sourceforge.net/projects/freenas/files/stable/0.7/FreeNAS-%1-embedded-%2.img/download").arg(cpuarch, relname), QString("%1ubninit").arg(targetPath));
}
}

1
2
3
4
5
6

//distrover.cpp
distroselect->addItem("FreeNAS", (QStringList() << "0.7.4919" <<
unetbootin::tr("Homepage: \"http://freenas.org/\">http://www.freenas.org
"
"Description: FreeNAS is an embedded open source NAS (Network-Attached Storage) distribution based on FreeBSD.
"
"Install Notes: The LiveCD version creates a RAM drive for FreeNAS, and uses a FAT formatted floppy disk or USB key for saving the configuration file. The embedded version allows installation to hard disk.") <<
"0.7.4919" << "0.7.4919_x64" << "0.7.1.5024_Live" << "0.7.1.4997_Live_x64"));

Segue:
I'm actually considering forking the unetbootin project to add support for a master distro list which can be updated remotely eliminating the requirement for users to download a new copy of the program if they wish to get the latest version of the list of pre-configured distros.

This has a little bit to do with the fact that I'll be required to take a few C++ courses at the University of Wyoming since Java is the standard language taught at the University of Arizona while I was there and I've never used C++ before. Can't be that hard right?

http://unetbootin.sourceforge.net/

Quote of the Day and Something Cool

bemasher — Fri, 11 Sep 2009 03:19:47 +0000

So I was sitting in my discrete structures analysis class today when a student asked a question about the homework. It went a little like this:

Student: How should we format pseudo-code in the homework?
Professor: Ah see pseudo-code is in that grey area. It's somewhere in between code and english.
Professor: You see me do it one way in class, the book does it another way and the homework assignment does it an even different way.
Professor: Think of it this way, pseudo-code is a lot like pornography: you'll know it when you see it.
Professor: So I'm not very worried about how you do your pseudo-code as long as I understand it.

The other cool thing that happened today was that I found out the owner of my favorite coffee shop is a pro-gun person. Apparently some of his relatives own a gun shop and he's done some shooting competitions. All very cool. It definitely explains why none of the employees ever freak out about me carrying my 1911 which looks giant on my hip compared to others. I'm a tall skinny guy if you didn't already know.