[bitcoin-dev] [BIP Draft] Datastream compression of Blocks and Transactions

From: Peter Tschipper <peter.tschipper@gmail•com>
To: gmaxwell@gmail•com, Bitcoin Dev <bitcoin-dev@lists•linuxfoundation.org>
Subject: [bitcoin-dev] [BIP Draft] Datastream compression of Blocks and Transactions
Date: Mon, 30 Nov 2015 15:12:24 -0800	[thread overview]
Message-ID: <565CD7D8.3070102@gmail.com> (raw)

@gmaxwell Bip Editor, and the Bitcoin Dev Community,

After several weeks of experimenting and testing with various
compression libraries I think there is enough evidence to show that
compressing blocks and transactions is not only beneficial in reducing
network bandwidth but is also provides a small performance boost when
there is latency on the network.

The following is a BIP Draft document for your review. 
(The alignment of the columns in the tables doesn't come out looking
right in this email but if you cut and paste into a text document they
are just fine)

<pre>
  BIP: ?
  Title: Datastream compression of Blocks and Tx's
  Author: Peter Tschipper <peter.tschipper@gmail•com>
  Status: Draft
  Type: Standards Track
  Created: 2015-11-30
</pre>

==Abstract==

To compress blocks and transactions, and to concatenate them together
when possible, before sending.

==Motivation==

Bandwidth is an issue for users that run nodes in regions where
bandwidth is expensive and subject to caps, in addition network latency
in some regions can also be quite high. By compressing data we can
reduce daily bandwidth used in a significant way while at the same time
speed up the transmission of data throughout the network. This should
encourage users to keep their nodes running longer and allow for more
peer connections with less need for bandwidth throttling and in
addition, may also encourage users in areas of marginal internet
connectivity to run nodes where in the past they would not have been
able to.

==Specification==

Advertise compression using a service bit.  Both peers must have
compression turned on in order for data to be compressed, sent, and
decompressed.

Blocks will be sent compressed.

Transactions will be sent compressed with the exception of those less
than 500 bytes.

Blocks will be concatenated when possible.

Transactions will be concatenated when possible or when a
MSG_FILTERED_BLOCK is requested.

Compression levels to be specified in "bitcoin.conf".

Compression and decompression can be completely turned off.

Although unlikely, if compression should fail then data will be sent
uncompressed.

The code for compressing and decompressing will be located in class
CDataStream.

Compression library LZO1x will be used.

==Rationale==

By using a service bit, compression and decompression can be turned
on/off completely at both ends with a simple configuration setting. It
is important to be able to easily turn off compression/decompression as
a fall back mechanism.  Using a service bit also makes the code fully
compatible with any node that does not currently support compression. A
node that do not present the correct service bit will simply receive
data in standard uncompressed format.

All blocks will be compressed. Even small blocks have been found to
benefit from compression.

Multiple block requests that are in queue will be concatenated together
when possible to increase compressibility of smaller blocks.
Concatenation will happen only if there are multiple block requests from
the same remote peer.  For example, if peer1 is requesting two blocks
and they are both in queue then those two blocks will be concatenated.
However, if peer1 is requesting 1 block and peer2 also one block, and
they are both in queue, then each peer is sent only its block and no
concatenation will occur. Up to 16 blocks (the max blocks in flight) can
be concatenated but not exceeding the MAX_PROTOCOL_MESSAGE_LENGTH.
Concatenated blocks compress better and further reduce bandwidth.

Transactions below 500 bytes do not compress well and will be sent
uncompressed unless they can be concatenated (see Table 3).

Multiple transaction requests that are in queue will be concatenated
when possible.  This further reduces bandwidth needs and speeds the
transfer of large requests for many transactions, such as with
MSG_FILTERED_BLOCK requests, or when the system gets busy and is flooded
with transactions.  Concatenation happens in the same way as for blocks,
described above.

By allowing for differing compression levels which can be specified in
the bitcoin.conf file, a node operator can tailor their compression to a
level suitable for their system.

Although unlikely, if compression fails for any reason then blocks and
transactions will be sent uncompressed.  Therefore, even with
compression turned on, a node will be able to handle both compressed and
uncompressed data from another peer.

By Abstracting the compression/decompression code into class
"CDataStream", compression can be easily applied to any datastream.

The compression library LZO1x-1 does not compress to the extent that
Zlib does but it is clearly the better performer (particularly as file
sizes get larger), while at the same time providing very good
compression (see Tables 1 and 2).  Furthermore, LZO1x-999 can provide
and almost Zlib like compression for those who wish to have more
compression, although at a cost.

==Test Results==

With the LZO library, current test results show up to a 20% compression
using LZO1x-1 and up to 27% when using LZO1x-999.  In addition there is
a marked performance improvement when there is latency on the network.
From the test results, with a latency of 60ms there is an almost 30%
improvement in performance when comparing LZO1x-1 compressed blocks with
uncompressed blocks (see Table 5).

The following table shows the percentage that blocks were compressed,
using two different Zlib and LZO1x compression level settings.

TABLE 1:
range = data size range
range           Zlib-1  Zlib-6  LZO1x-1 LZO1x-999
-----------     ------  ------  ------- --------
0-250           12.44   12.86   10.79   14.34
250-500         19.33   12.97   10.34   11.11   
600-700         16.72   n/a     12.91   17.25
700-800         6.37    7.65    4.83    8.07
900-1KB         6.54    6.95    5.64    7.9
1KB-10KB        25.08   25.65   21.21   22.65
10KB-100KB      19.77   21.57   4.37    19.02
100KB-200KB     21.49   23.56   15.37   21.55
200KB-300KB     23.66   24.18   16.91   22.76
300KB-400KB     23.4    23.7    16.5    21.38
400KB-500KB     24.6    24.85   17.56   22.43
500KB-600KB     25.51   26.55   18.51   23.4
600KB-700KB     27.25   28.41   19.91   25.46
700KB-800KB     27.58   29.18   20.26   27.17
800KB-900KB     27      29.11   20      27.4
900KB-1MB       28.19   29.38   21.15   26.43
1MB -2MB        27.41   29.46   21.33   27.73

The following table shows the time in seconds that a block of data takes
to compress using different compression levels.  One can clearly see
that LZO1x-1 is the fastest and is not as affected when data sizes get
larger.

TABLE 2:
range = data size range
range           Zlib-1  Zlib-6  LZO1x-1 LZO1x-999
-----------     ------  ------  ------- ---------
0-250           0.001   0       0       0
250-500         0       0       0       0.001
500-1KB         0       0       0       0.001
1KB-10KB        0.001   0.001   0       0.002
10KB-100KB      0.004   0.006   0.001   0.017
100KB-200KB     0.012   0.017   0.002   0.054
200KB-300KB     0.018   0.024   0.003   0.087
300KB-400KB     0.022   0.03    0.003   0.121
400KB-500KB     0.027   0.037   0.004   0.151
500KB-600KB     0.031   0.044   0.004   0.184
600KB-700KB     0.035   0.051   0.006   0.211
700KB-800KB     0.039   0.057   0.006   0.243
800KB-900KB     0.045   0.064   0.006   0.27
900KB-1MB       0.049   0.072   0.006   0.307

TABLE 3:
Compression of Transactions (without concatenation)
range = block size range
ubytes = average size of uncompressed transactions
cbytes = average size of compressed transactions
cmp% = the percentage amount that the transaction was compressed
datapoints = number of datapoints taken

range       ubytes    cbytes    cmp%    datapoints
----------  ------    ------    ------  ----------    
0-250       220       227       -3.16   23780
250-500     356       354       0.68    20882
500-600     534       505       5.29    2772
600-700     653       608       6.95    1853
700-800     757       649       14.22   578
800-900     822       758       7.77    661
900-1KB     954       862       9.69    906
1KB-10KB    2698      2222      17.64   3370
10KB-100KB  15463     12092     21.80   15429

The above table shows that transactions don't compress well below 500
bytes but do very well beyond 1KB where there are a great deal of those
large spam type transactions.   However, most transactions happen to be
in the < 500 byte range.  So the next step was to appy concatenation for
those smaller transactions.  Doing that yielded some very good
compression results.  Some examples as follows:

The best one that was seen was when 175 transactions were concatenated
before being compressed.  That yielded a 20% compression ratio, but that
doesn't take into account the savings from the unneeded 174 message
headers (24 bytes each) as well as 174 TCP ACKs of 52 bytes each which
yields and additional 76*174 = 13224 byte savings, making for an overall
bandwidth savings of 32%:

     2015-11-18 01:09:09.002061 compressed data from 79890 to 67426
txcount:175

However, that was an extreme example.  Most transaction aggregates were
in the 2 to 10 transaction range.  Such as the following:

     2015-11-17 21:08:28.469313 compressed data from 3199 to 2876 txcount:10

But even here the savings of 10% was far better than the "nothing" we
would get without concatenation, but add to that the 76 byte * 9
transaction savings and we have a total 20% savings in bandwidth for
transactions that otherwise would not be compressible.  Therefore the
concatenation of small transactions can also save bandwidth and speed up
the transmission of those transactions through the network while keeping
network and message queue chatter to a minimum.

==Choice of Compression library==

LZO was chosen over Zlib.  LZO is the fastest most scalable option when
used at the lowest compression setting which will be a performance boost
for users that prefer performance over bandwidth savings. And at the
higher end, LZO provides good compression (although at a higher cost)
which approaches that of Zlib.

Other compression libraries investigated were Snappy, LZOf, fastZlib and
LZ4 however none of these were found to be suitable, either because they
were not portable, lacked the flexibility to set compression levels or
did not provide a useful compression ratio.

The following two tables show results in seconds for syncing the first
200,000 blocks. Tests were run on a high-speed wireless LAN with very
little latency, and also run with a 60ms latency which was induced with
"Netbalancer".

TABLE 4:
Results shown in seconds on highspeed wireless LAN (no induced latency)
Num blks sync'd  Uncmp  Zlib-1  Zlib-6  LZO1x-1  LZO1x-999
---------------  -----  ------  ------  -------  ---------
10000            255    232     233     231      257      
20000            464    414     420     407      453      
30000            677    594     611     585      650      
40000            887    787     795     760      849     
50000            1099   961     977     933      1048   
60000            1310   1145    1167    1110     1259  
70000            1512   1330    1362    1291     1470  
80000            1714   1519    1552    1469     1679   
90000            1917   1707    1747    1650     1882  
100000           2122   1905    1950    1843     2111    
110000           2333   2107    2151    2038     2329  
120000           2560   2333    2376    2256     2580   
130000           2835   2656    2679    2558     2921 
140000           3274   3259    3161    3051     3466   
150000           3662   3793    3547    3440     3919   
160000           4040   4172    3937    3767     4416   
170000           4425   4625    4379    4215     4958   
180000           4860   5149    4895    4781     5560    
190000           5855   6160    5898    5805     6557    
200000           7004   7234    7051    6983     7770   

TABLE 5:
Results shown in seconds with 60ms of induced latency
Num blks sync'd  Uncmp  Zlib-1  Zlib-6  LZO1x-1  LZO1x-999
---------------  -----  ------  ------  -------  ---------
10000            219    299     296     294      291
20000            432    568     565     558      548
30000            652    835     836     819      811
40000            866    1106    1107    1081     1071
50000            1082   1372    1381    1341     1333
60000            1309   1644    1654    1605     1600
70000            1535   1917    1936    1873     1875
80000            1762   2191    2210    2141     2141
90000            1992   2463    2486    2411     2411
100000           2257   2748    2780    2694     2697
110000           2627   3034    3076    2970     2983
120000           3226   3416    3397    3266     3302
130000           4010   3983    3773    3625     3703
140000           4914   4503    4292    4127     4287
150000           5806   4928    4719    4529     4821
160000           6674   5249    5164    4840     5314
170000           7563   5603    5669    5289     6002
180000           8477   6054    6268    5858     6638
190000           9843   7085    7278    6868     7679
200000           11338  8215    8433    8044     8795

==Backward compatibility==

Being unable to present the correct service bit, older clients will
continue to receive standard uncompressed data and will be fully
compatible with this change.

==Fallback==

It is important to be able to entirely and easily turn off compression
and decompression as a fall back mechanism. This can be done with a
simple bitcoin.conf setting of "compressionlevel=0". Only one of the two
connected peers need to set compressionlevel=0 in order to turn off
compression and decompression completely.

==Deployment==

This enhancement does not require a hard or soft fork.

==Service Bit==

During the testing of this implementation, service bit 28 was used,
however this enhancement will require a permanently assigned service bit.

==Implementation==

This implementation depends on the LZO compression library: lzo-2.09

     https://github.com/ptschip/bitcoin/tree/compress

==Copyright==

This document is placed in the public domain.