Recherche avancée

Sur d’autres sites (9862)

• Processing Big Data Problems

  8 janvier 2011, par Multimedia Mike — Big Data

  I’m becoming more interested in big data problems, i.e., extracting useful information out of absurdly sized sets of input data. I know it’s a growing field and there is a lot to read on the subject. But you know how I roll— just think of a problem to solve and dive right in.

  Here’s how my adventure unfolded.

  The Corpus
  I need to run a command line program on a set of files I have collected. This corpus is on the order of 350,000 files. The files range from 7 bytes to 175 MB. Combined, they occupy around 164 GB of storage space.

  Oh, and said storage space resides on an external, USB 2.0-connected hard drive. Stop laughing.

  A file is named according to the SHA-1 hash of its data. The files are organized in a directory hierarchy according to the first 6 hex digits of the SHA-1 hash (e.g., a file named a4d5832f... is stored in a4/d5/83/a4d5832f...). All of this file hash, path, and size information is stored in an SQLite database.

  First Pass
  I wrote a Python script that read all the filenames from the database, fed them into a pool of worker processes using Python’s multiprocessing module, and wrote some resulting data for each file back to the SQLite database. My Eee PC has a single-core, hyperthreaded Atom which presents 2 CPUs to the system. Thus, 2 worker threads crunched the corpus. It took awhile. It took somewhere on the order of 9 or 10 or maybe even 12 hours. It took long enough that I’m in no hurry to re-run the test and get more precise numbers.

  At least I extracted my initial set of data from the corpus. Or did I ?

  Think About The Future
  
  A few days later, I went back to revisit the data only to notice that the SQLite database was corrupted. To add insult to that bit of injury, the script I had written to process the data was also completely corrupted (overwritten with something unrelated to Python code). BTW, this is was on a RAID brick configured for redundancy. So that’s strike 3 in my personal dealings with RAID technology.

  I moved the corpus to a different external drive and also verified the files after writing (easy to do since I already had the SHA-1 hashes on record).

  The corrupted script was pretty simple to rewrite, even a little better than before. Then I got to re-run it. However, this run was on a faster machine, a hyperthreaded, quad-core beast that exposes 8 CPUs to the system. The reason I wasn’t too concerned about the poor performance with my Eee PC is that I knew I was going to be able to run in on this monster later.

  So I let the rewritten script rip. The script gave me little updates regarding its progress. As it did so, I ran some rough calculations and realized that it wasn’t predicted to finish much sooner than it would have if I were running it on the Eee PC.

  Limiting Factors
  It had been suggested to me that I/O bandwidth of the external USB drive might be a limiting factor. This is when I started to take that idea very seriously.

  The first idea I had was to move the SQLite database to a different drive. The script records data to the database for every file processed, though it only commits once every 100 UPDATEs, so at least it’s not constantly syncing the disc. I ran before and after tests with a small subset of the corpus and noticed a substantial speedup thanks to this policy chance.

  Then I remembered hearing something about "atime" which is access time. Linux filesystems, per default, record the time that a file was last accessed. You can watch this in action by running 'stat <file> ; cat <file> > /dev/null ; stat <file>' and observe that the "Access" field has been updated to NOW(). This also means that every single file that gets read from the external drive still causes an additional write. To avoid this, I started mounting the external drive with '-o noatime' which instructs Linux not to record "last accessed" time for files.

  On the limited subset test, this more than doubled script performance. I then wondered about mounting the external drive as read-only. This had the same performance as noatime. I thought about using both options together but verified that access times are not updated for a read-only filesystem.

  A Note On Profiling
  Once you start accessing files in Linux, those files start getting cached in RAM. Thus, if you profile, say, reading a gigabyte file from a disk and get 31 MB/sec, and then repeat the same test, you’re likely to see the test complete instantaneously. That’s because the file is already sitting in memory, cached. This is useful in general application use, but not if you’re trying to profile disk performance.

  Thus, in between runs, do (as root) 'sync; echo 3 > /proc/sys/vm/drop_caches' in order to wipe caches (explained here).

  Even Better ?
  I re-ran the test using these little improvements. Now it takes somewhere around 5 or 6 hours to run.

  I contrived an artificially large file on the external drive and did some 'dd' tests to measure what the drive could really do. The drive consistently measured a bit over 31 MB/sec. If I could read and process the data at 30 MB/sec, the script would be done in about 95 minutes.

  But it’s probably rather unreasonable to expect that kind of transfer rate for lots of smaller files scattered around a filesystem. However, it can’t be that helpful to have 8 different processes constantly asking the HD for 8 different files at any one time.

  So I wrote a script called stream-corpus.py which simply fetched all the filenames from the database and loaded the contents of each in turn, leaving the data to be garbage-collected at Python’s leisure. This test completed in 174 minutes, just shy of 3 hours. I computed an average read speed of around 17 MB/sec.

  Single-Reader Script
  I began to theorize that if I only have one thread reading, performance should improve greatly. To test this hypothesis without having to do a lot of extra work, I cleared the caches and ran stream-corpus.py until 'top' reported that about half of the real memory had been filled with data. Then I let the main processing script loose on the data. As both scripts were using sorted lists of files, they iterated over the filenames in the same order.

  Result : The processing script tore through the files that had obviously been cached thanks to stream-corpus.py, degrading drastically once it had caught up to the streaming script.

  Thus, I was incented to reorganize the processing script just slightly. Now, there is a reader thread which reads each file and stuffs the name of the file into an IPC queue that one of the worker threads can pick up and process. Note that no file data is exchanged between threads. No need— the operating system is already implicitly holding onto the file data, waiting in case someone asks for it again before something needs that bit of RAM. Technically, this approach accesses each file multiple times. But it makes little practical difference thanks to caching.

  Result : About 183 minutes to process the complete corpus (which works out to a little over 16 MB/sec).

  Why Multiprocess
  Is it even worthwhile to bother multithreading this operation ? Monitoring the whole operation via 'top', most instances of the processing script are barely using any CPU time. Indeed, it’s likely that only one of the worker threads is doing any work most of the time, pulling a file out of the IPC queue as soon the reader thread triggers its load into cache. Right now, the processing is usually pretty quick. There are cases where the processing (external program) might hang (one of the reasons I’m running this project is to find those cases) ; the multiprocessing architecture at least allows other processes to take over until a hanging process is timed out and killed by its monitoring process.

  Further, the processing is pretty simple now but is likely to get more intensive in future iterations. Plus, there’s the possibility that I might move everything onto a more appropriately-connected storage medium which should help alleviate the bottleneck bravely battled in this post.

  There’s also the theoretical possibility that the reader thread could read too far ahead of the processing threads. Obviously, that’s not too much of an issue in the current setup. But to guard against it, the processes could share a variable that tracks the total number of bytes that have been processed. The reader thread adds filesizes to the count while the processing threads subtract file sizes. The reader thread would delay reading more if the number got above a certain threshold.

  Leftovers
  I wondered if the order of accessing the files mattered. I didn’t write them to the drive in any special order. The drive is formatted with Linux ext3. I ran stream-corpus.py on all the filenames sorted by filename (remember the SHA-1 naming convention described above) and also by sorting them randomly.

  Result : It helps immensely for the filenames to be sorted. The sorted variant was a little more than twice as fast as the random variant. Maybe it has to do with accessing all the files in a single directory before moving onto another directory.

  Further, I have long been under the impression that the best read speed you can expect from USB 2.0 was 27 Mbytes/sec (even though 480 Mbit/sec is bandied about in relation to the spec). This comes from profiling I performed with an external enclosure that supports both USB 2.0 and FireWire-400 (and eSata). FW-400 was able to read the same file at nearly 40 Mbytes/sec that USB 2.0 could only read at 27 Mbytes/sec. Other sources I have read corroborate this number. But this test (using different hardware), achieved over 31 Mbytes/sec.

• Révision 17381 : squeletiser l’affichage des auteurs en ligne, ça sera plus facile a personaliser.

  12 mars 2011, par cedric -

  Au passage on evacue le lien messagerie (disponible le cas echeant sur la fiche auteur linkee), et on limite au 10 derniers auteurs connectes (par ordre chronologique inverse), en indiquant simplement le nombre total si il y en a plus que (...)

• Creating A Lossless SMC Encoder

  26 avril 2011, par Multimedia Mike — General

  Look, I can’t explain how or why I come up with this stuff. For some reason, I thought it would be interesting to write a new encoder for the Apple SMC video codec. I can’t even remember why. I just sat down the other day, started writing, and now I have a lossless SMC encoder that I’m not sure what to do with. Maybe this is to be my new thing— writing encoders for marginal multimedia formats.

  Introduction
  SMC is a vector quantizer (a lossy method) but I decided to attack it from the angle of lossless encoding. A.k.a. Apple Graphics Codec, SMC operates on 4x4 blocks in an 8-bit paletted colorspace. Each 4x4 block can be encoded with 1, 2, 4, 8, or 16 colors. Blocks can also be skipped (copied from previous frame) or copied from blocks rendered immediately prior within the same frame.

  Step 1 : Validating Infrastructure
  The goal of this step is to encode the most braindead SMC frame possible and see if FFmpeg/libav’s QuickTime muxer can create a valid file. I think the simplest frame would be one in which each vector is encoded with the single-color mode, starting with color 0 and incrementing through the palette.

  Status : Successful. The only ’trick’ was to set avctx->bits_per_coded_sample to 8. (For fun, this can also be set to 40 (8 | 0x20) to specify a grayscale palette.)
  

  
  
  

  Step 2 : Preprocessing
  The video frames will arrive at the encoder as 32-bit RGB. These will need to be converted to a paletted colorspace before encoding. I don’t want to use FFmpeg’s default dithering approach as this will result in a substantial loss of quality as described in this post. I would rather maintain a palette built from observed colors throughout successive frames. If the total number of unique observed colors ever exceeds 256, error out.

  That’s what I would like to do. However, I noticed that FFmpeg/libav’s QuickTime muxer has never taken into account the possibility of encoding palettes. The path of least resistance in this case is to dither the input to match QuickTime’s default 8-bit palette (if a paletted QuickTime file does not specify a palette, a default 1-, 2-, 4-, or 8-bit palette is selected).

  Status : Successful, if slow. I definitely need to optimize this step later.

  Step 3 : Most Naive Encoding
  The most basic encoding is to "encode" each block as a 16-color block. This will actually result in a slightly larger frame size than a raw encoding since each 4x4 block will be prepended by a byte opcode (0xE0 in this case) to indicate encoding mode. This should demonstrate that the encoder is functioning at the most basic level.

  Status : Successful. Try not to laugh too hard at the Big Buck Bunny dithered to an 8-bit palette :

  
  
  

  Step 4 : Better Representation
  It seems to me that encoding this format (losslessly) will entail performing vector operations on lots of 16-element (4x4-pixel) vectors. These could be done on the frame as-is, but it strikes me as more efficient and perhaps less error prone to rearrange the input images into a vector of vectors (or array of arrays if you prefer) :

    0  1  2  3  w ...
    4  5  6  7  x ...
    8  9  A  B  y ...
    C  D  E  F  z ...
  

    0 : [0 1 2 3 4 5 6 7 8 9 A B C D E F]
    1 : [...]
  

  Status : Successful.

  Step 5 : Add Interframe Skip Codes
  Time to add a bit of brainpower to the proceedings : On non-keyframes, compare the current vector to the vector at the same position from the previous frame.

  Test this by encoding a pair of identical frames. Ideally, all codes should be skip codes.

  Status : Successful, though my vector matching function could probably be improved.

  Step 6 : Analyze Blocks For Optimal Color Coding
  This is where things get potentially interesting, algorithmically. At least, I need to figure out (or look up) an algorithm to count the unique elements in a vector.

  Naive algorithm (i.e., first thing I can think of) :

  • initialize a count variable to 0
  • initialize an array of 256 flags to false
  • for each 8-bit element in vector :
    • if flag array[element] is 0, set array[element] to true and increment count

  Status : Successful. Here is the distribution for the 640x360 Big Buck Bunny title :

  1194 4636 4113 2140 1138 568 325 154 80 36 9 5 2 0 0 0

  Or, in pretty graph form, demonstrating that vectors with few distinct elements dominate :
  

  
  
  

  Step 7 : Encode Monochrome Blocks
  At this point, the structure is starting to come together pretty well. This phase involves encoding a 0x60 opcode and a palette index when the count_distinct() function returns 1.

  Status : Absolutely no problem.

  Step 8 : Encode 2-, 4-, and 8-color Modes
  This step is a little more involved. This is where SMC’s 2-, 4-, and 8-color circular palette caches come into play. E.g., when the first 2-color block is encoded, the pair of colors it uses will be inserted into entry 0 of the 2-color cache. During the next 2-color block encoding, if the block uses a pair of colors that already occurs in the cache, the encoding can reference that cache entry. Otherwise, it adds the pair to the next available cache entry, looping back around to 0 as necessary.

  I think I should modify the count_distinct() function to also return a 16-byte array that contains a sorted list of the palette indicies used in the vector. The color pair cache will contain 256 16-bit, 32-bit ints for the quads and 64-bit ints for the octets. This will allow a slightly faster linear cache search.

  Status : The 2-color encoding wasn’t too much trouble and I was able to adapt it to the 4-color mode pretty quickly afterward. I’m still having trouble with the insane 8-color coding mode, though. So that’s commented out for the time being.

  Step 9 : Run Encoding and Putting It All Together
  For each frame, convert the input pixels to a paletted format via one method or another (match to default QuickTime palette for first pass). Then, preprocess each vector to determine the minimum number of elements that can be used to represent it, storing the sorted list of distinct colors in a separate array. The number of elements can either be 0 (only for interframes and indicates a skip block), 1, 2, 4, 8, or 16. Also during this phase, for each vector after the first, test if the vector is the same as the previous vector. If it is, denote this fact in the preprocessed encoding (set the high bit of the element count number).

  Finally, pack it into the bytestream. Iterate through the element count array and search for the longest runs of elements that are encoded with the same mode (up to 256 for skip modes, up to 16 for other modes). If the high bit of an element count is set, that indicates that a copy mode can be encoded. Look for the longest run of element counts with the high bit set and encode a copy mode.

  Status : In-process. Will finish this as motivation strikes.