Date: Thu, 31 Jul 2008 18:56:03 +1000 From: "Andrew Hill" <lists@thefrog.net> To: freebsd-java@freebsd.org Subject: Re: Azureus + 7-STABLE == Slow download + No Upload Message-ID: <16a6ef710807310156xec3f034xe1c15af16dd074c1@mail.gmail.com> In-Reply-To: <f1f18921-a6f7-4d85-a52a-31a10a3bedf5@u6g2000prc.googlegroups.com> References: <ft0c03$2jc1$1@FreeBSD.csie.nctu.edu.tw> <f1f18921-a6f7-4d85-a52a-31a10a3bedf5@u6g2000prc.googlegroups.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Apr 3, 2:24 am, scra...@hub.org ("Marc G. Fournier") wrote:
> > > The reason that we've persisted with doing patchsets for Java is so
> that
> > > everybody can get at the source and investigate/fix things which are
> > > important to them. I've persistently argued against switching solely
> to
> > > binary releases, when that has come up, for just this reason.
> >
> > First, a note ... I think the Java group is doing a fantastic job,
> > considering all of the hurdle that you guys had to jump through at the
> > start of this with NDA (I believe?) and such ... please do not take *any*
> > of what I have (or will) say as anything but an attempt to further
> > improve, as they aren't meant as anything but ...
> >
> > > So, if you can't find someone on the porting team who is both
> interested
> > > and has time to investigate it right now then there are a number of
> things
> > > you can do yourrselves to start the investigation.
> >
> > In this case, I'm curious as to why nobody is taking this as an
> > opportunity ...Azureusis a huge piece of software, that tends to make
> > extensive use of threading (from what I can tell), and since at least
> 6.x,
> > has had problem reports when trying to do 'two things at once' ... if
> this
> > is a coding bug withAzureusitself, fine, but if this is a bug with
> > FreeBSD (ie. our threading + java) thatAzureushighlights to the extreme,
> > then if we can getAzureusfixed, that would trickle down to other less
> > 'intense' java applications, and, possible even non-java ones ...
> >
> > *But* ... after installing deluge (python), which I'm getting massive
> > transfer rates (such that I'd never seen withAzureuseven at its best)
> > even with 20+ torrents on the go, I don't see it as being a FreeBSD
> > threading issue, except for how that interacts with Java+Azureus...
> >
> > Now, it could be like with Wine, whereAzureuswas programmed with 'linux
> > in mind', so that someone is missing / broken in FreeBSD/java in relation
> > ... but even knowing *that* would at least make it easier for ppl posting
> > about such issues ...
> >
> > > . Look through the list archives into what people have found out about
> > > problems withAzureusbefore. This isn't the first time this has come
> > > up.
> >
> > I've done that, and you are right, it isn't the firs time ... the bulk of
> > the reports have revolved around 6.x, where you need to add an entry to
> > /etc/libmap.conf to map pthread -> thr ... with that,Azureuson 6.x works
> > fine, but with 7.x, thr is the default library, and I've checked jdk15
> > using ldd to make sure that that is what is being used ...
> >
> > The only other issue I've found relates to IPv6, but I don't have that
> > built into my kernel, and, just in case, I did add the flag that was
> > suggested to force IPv4 ...
> >
> > The other email's that I've seen on the subject generally went along the
> > lines of "I gave up onAzureus+Java and moved to something else to fix the
> > problem" ...
> >
> > > . Look at what the kernel threads are doing periodically with ps(1).
> >
> > What should one be looking at here? I have no problems with doing the
> > 'leg work' on this, just need to know what I need to be looking for ...
> >
> > > . Dump the Java threads periodically.
> >
> > Huh?
> >
> > > . RunAzureusunder ktrace and look at what syscalls are being made.
> > > There is going to be a lot of output from ktrace I expect, so its
> > > going to take some time to sort through.
> >
> > I have the disk space, but insufficicent knowledge to know what I'm
> > looking for from the output ...
> >
> > > . Run this on 6.x and 7.x and try the different threading libraries.
> >
> > 6.x runs fine with the pthread->thr map in /etc/libmap.conf ... I've
> tried
> > it on 7.x with both diablo (uses pthreads, I believe) and jdk15 (with
> thr)
> > .. same results ...
> >
> > > . Run it on Linux and compare what it does if you're interested.
> >
> > Not an option ... I don't have a Linux box to run it on ...
> >
> > > . Run it under different JDKs to see if that makes any difference.
> >
> > Tried both diablo and jdk15 ... haven't tried the linux jdk though, will
> > look at installing that tonight ...
> >
>
This thread's gone a bit quiet, so I don't know if there's a solution I've
missed (or more likely everyone's given up on Azureus for deluge), but I
just tried azureus again to see if it'd been fixed in the last couple of
months and I saw the same problems - plenty of users to connect to, plenty
of fully established connections, but very few peers/seeds were able to
send/receive data - I noticed that it wasn't that only Azureus users could
send, but a very select few of the Azureus users...
After enabling a few extra columns in the peers display (most notably the
'Encryption' column) I saw that it was only the users with 'RC4-160 (UDP)'
in the encryption column that were getting data through. It's not just
encryption alone, because those with just 'RC4-160' weren't sending any
data, it seems to be specifically those with the 'UDP' suffix, which I'm
assuming means they're connected over UDP rather than TCP.
I know people have been talking about threading as a likely culprit, but
perhaps its something to with TCP connections (or threading of TCP
connections specifically?) Unfortunately I'm not really sure how to go about
testing/confirming/fixing this, but hopefully this info is useful for
someone who's better equipped to find the problem...
On a side note, it seems that the Azureus/Vuze devs have decided to add in
an option to prefer UDP connections which might make this easier to test my
theory in the near future (see
http://forum.vuze.com/thread.jspa?threadID=76757)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?16a6ef710807310156xec3f034xe1c15af16dd074c1>