Re: How (and why) to use the Relay Network

[Matt Corallo]

There were a few major updates and both clients should update...The java client had a rather dumb bug that prevented it from relying outgoing blocks, though it received them fine...the python client was, as is all python, so horrendously non-performant that even a network rtt for a inv->getdata->block process to be faster. With some minor effort the python client now should reliably beat network rtts (seriously python???), though I'd advise using the java client where possible.

[PatMan]

Damn that java - I just uninstalled it too........

[Matt Corallo]

Damn that java - I just uninstalled it too........

Well, if you really wanted the python version you could use Jython, but then you're stuck with java anyway...I may benchmark pypy later and see if it helps but early testing shows its ~as fast.

[PatMan]

Really? That surprises me, I've never used pypy but have heard it's much faster than std python.......

[Matt Corallo]

Really? That surprises me, I've never used pypy but have heard it's much faster than std python.......

Well, its a JIT, and my tests may simply not have been long-running enough to give it sufficient warm-up time.

[spiccioli]

Really? That surprises me, I've never used pypy but have heard it's much faster than std python.......

Well, its a JIT, and my tests may simply not have been long-running enough to give it sufficient warm-up time.

Hi,

I came to the same conclusions running p2pool long ago on pypy, it is as fast as python.

See here, for example.

spiccioli

[-ck]

There were a few major updates and both clients should update...The java client had a rather dumb bug that prevented it from relying outgoing blocks, though it received them fine...the python client was, as is all python, so horrendously non-performant that even a network rtt for a inv->getdata->block process to be faster. With some minor effort the python client now should reliably beat network rtts (seriously python???), though I'd advise using the java client where possible.

Upgraded, thanks. I did notice messages in both directions now instead of before on the java client.

[Matt Corallo]

There is now a C++ client. Those who had been using the python client should probably switch to it (its semantics are identical, it runs on Linux and under Wine, so I assume Windows) and it is much, much faster than the python client.

[PatMan]

Great stuff Matt! Testing now.......

[jedimstr]

There is now a C++ client. Those who had been using the python client should probably switch to it (its semantics are identical, it runs on Linux and under Wine, so I assume Windows) and it is much, much faster than the python client.

Any chance you or someone else could port/compile/brew a Mac C++ version?

I tried porting and compiling myself but all the includes FUBAR'd the attempt. Or, I just need more coffee before I try again.

Been running the Java client for awhile now, but can we get the jar added to the Git Repository too alongside the other clients? It'll make keeping up-to-date much easier.

[Matt Corallo]

Any chance you or someone else could port/compile/brew a Mac C++ version?

I tried porting and compiling myself but all the includes FUBAR'd the attempt. Or, I just need more coffee before I try again.

Hmm, yea, I assume one or more of the socket calls isnt right? I didnt try it on any BSD. Also, it needs C++11, so if the version of clang OSX ships is particularly out-of-date it may fail there too...I'm happy to help test it if you can get me a list of compile errors.

[Matt Corallo]

Been running the Java client for awhile now, but can we get the jar added to the Git Repository too alongside the other clients? It'll make keeping up-to-date much easier.

Done.

[DrHaribo]

Is there a maven repo I need to add to build the java client? It's not finding the required bitcoinj version.

Code:

Missing:
----------
1) com.google:bitcoinj:jar:0.12-SNAPSHOT_nodelay

[Matt Corallo]

Is there a maven repo I need to add to build the java client? It's not finding the required bitcoinj version.

Code:

Missing:
----------
1) com.google:bitcoinj:jar:0.12-SNAPSHOT_nodelay

If you're building the java client direct from source, it wants the branch at https://github.com/TheBlueMatt/bitcoinj/tree/nodelay. The only real difference is it sets TCP_NODELAY.

[IYFTech]

Hi Matt,

Sorry to go a bit off topic, but I was looking at your git repo & noticed you've worked on TextSecure. I've been using it for several weeks now - it's great work! Well done & kudos to you 

Sorry guys, carry on.......

[DrHaribo]

If you're building the java client direct from source, it wants the branch at https://github.com/TheBlueMatt/bitcoinj/tree/nodelay. The only real difference is it sets TCP_NODELAY.

Got it. Had to add _nodelay to the version in core/pom.xml so the RelayClient would find it after "mvn install". Strangely the tests for bitcoinj fail if you run them twice without "mvn clean". At least they did for me. But I guess that's a problem with the tests, not the library itself.

Perhaps it's an idea for pools to set up direct peering? Could shave off a little time.

Is it OK to pass new blocks directly to the RelayNodeClient instead of passing them through the local bitcoind ? Could shave off a little bit more time.

[Matt Corallo]

Got it. Had to add _nodelay to the version in core/pom.xml so the RelayClient would find it after "mvn install". Strangely the tests for bitcoinj fail if you run them twice without "mvn clean". At least they did for me. But I guess that's a problem with the tests, not the library itself.

Perhaps it's an idea for pools to set up direct peering? Could shave off a little time.

Is it OK to pass new blocks directly to the RelayNodeClient instead of passing them through the local bitcoind ? Could shave off a little bit more time.

If you used that branch it should already have had _nodelay added to its version number (and you should probably use that branch of your OS may hold onto data before relaying it for some time).

re: direct peering: sure, pools could do this (most already do afaiu).

Sure, no harm in passing a block out quickly, but this is also why you should treat blocks that come from the relay network identically as some arbitrary blocks you get from an unknown source on the p2p network. See, for example, the for-w branch at https://github.com/TheBlueMatt/RelayNode/tree/for-w/client which I whipped up for another poolop so that they could peer a single relay network client with a ton of local pool daemons and have it properly route blocks that come in from the relay network to their bitcoind and from one of their pool daemons to all the other pool daemons and the relay network all at once.

[DrHaribo]

If you used that branch it should already have had _nodelay added to its version number (and you should probably use that branch of your OS may hold onto data before relaying it for some time).

Sorry, I forgot to switch to the nodelay branch.

re: direct peering: sure, pools could do this (most already do afaiu).

Yeah, but I meant with the relay software, running a RelayNode instead of a RelayClientNode. Not sure how much there is to save there, but there may be something.

[norgan]

I've downloaded the exe client and am running it with "D:\Temp\relay\relaynetworkclient.exe public.au.relay.mattcorallo.com 127.0.0.1 8333" but I just get the following:

Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)
Closing bitcoind socket, failed to connect() (10049: Unknown error)

What am I doing wrong?

[Matt Corallo]

I've downloaded the exe client and am running it with "D:\Temp\relay\relaynetworkclient.exe public.au.relay.mattcorallo.com 127.0.0.1 8333" but I just get the following:

Closing bitcoind socket, failed to connect() (10049: Unknown error)
Closing relay socket, failed to connect() (10049: Unknown error)

What am I doing wrong?

10049 == WSAEADDRNOTAVAIL according to http://msdn.microsoft.com/en-us/library/windows/desktop/ms740668(v=vs.85).aspx. Any chance you're still on XP or have IPv6 explicitly disabled somewhere, which I think I might have broken in a recent push.