Very preliminary Bloom filter test results

[jim618]

In the last couple of days I have been upgrading the test MultiBit code to use the bloom filters that are coming in bitcoind v0.8.0 + bitcoinj v0.7. (These are still in test yet).
Here are the very preliminary test results I am seeing.

Bloom filter testing

Server configuration:
An old Linux VM, only 512 MB ram.
Software: bitcoind master (what will become bitcoind v0.8 )
Mike set this up for me - thanks!

Client machine 1:
Mac laptop. dual core, 2 GHz.
Network: 3G, about 80 KB/s down.

MultiBit - test code, using bitcoinj-0.7-SNAPSHOT + encrypted wallets.
MultiBit has 6 wallets open, about 80 private keys.
This machine normally gets about 1 block/s when syncing.

Client machine 2:
Centos 6 server, Athlon II quad core, 2.3 GHz.
Network: 100 MB/s

MultiBit - test code, using bitcoinj-0.7-SNAPSHOT + encrypted wallets.
MultiBit has 1 wallet open, 3 private keys.
I normally get about 10 block/s with this machine.
(This is the multibit.org machine so it is also running a website with variable load).


Test performed:
Download 1 month of blocks (using 'Tools | Replay blockchain and transactions')
This was 4366 to 4385 blocks.


Results using client machine 1 (Mac laptop):
Using 3G
Sync time: 198 seconds.
22 block/s

Notes: Yesterday I saw about 40 block/s with a better network connection.
Both CPUs busy, at 100% max over half the time.


Using coffee shop wifi (About 200 KB/s down).
Sync time: 120 seconds.
36 blocks/s

Notes: Both CPUS maxed out almost all the time.


Results using client machine 2 (multibit.org server):
Sync time: 274 secs.
16 blocks/s

Notes: I am a bit surprised this much faster machine comes in slower than a laptop, perhaps backend related ?

[1714793992]

1714793992

[1714793992]

1714793992

[1714793992]

1714793992

[1714793992]

1714793992

[1714793992]

1714793992

[1714793992]

1714793992

[Mike Hearn]

You should be able to get much faster than that on a laptop with coffee shop wifi. Profiling the app to find out where the CPU time is going would be the next step.

[jim618]

Yeah - I suspect that I am wasting a lot of CPU updating the UI dealing with events (unnecessarily).
It is only early days yet.

[jim618]

In the v0.5.8beta code that has just been released, the bloom filter speeds I am getting are:

Mac laptop, with 3G network: about 80 blocks/second.

Mac laptop, with WiFi (1 MB/s down, 100kB/s up): about 130 blocks per second.

[freemoney458]

I just tried out v0.5.8beta
on a MacBook running OS X version 10.5.8,
connected to 1 bitcoind v0.8rc1 peer at riker.plan99.net (courtesy of Mike Hearn).
My internet connection is running at 100 Mbit/s.

One wallet with 20 addresses, reloading the whole blockchain (220867 blocks) from the genesis block took less than 11 minutes.
That equals 335 blocks/second.

Congratulations and a big THANK YOU to all bloom filter developers!

[jim618]

That's great feedback Freemoney - thanks for that.