[prev in list] [next in list] [prev in thread] [next in thread] 

List:       intermezzo-discuss
Subject:    Re: [Fwd: InterMezzo success report (was Re: first crash)]
From:       Shirish Hemant Phatak <shirish () nustorage ! com>
Date:       2001-07-18 19:15:29
[Download RAW message or body]

Vladimir Ivaschenko wrote:

> Shirish Hemant Phatak wrote:
> 
> > Hi,
> > partition.
> > > > 
> > > 
> > > So, journal file is created automatically? Where? Intermezzo docs in doc/ still \
> > > state that journal file has to be created manually (Intermezzo-ext3.txt)
> > 
> > No, the journal file must be created manually (actually I recomend tune2fs -j).  \
> > When you first mount the fs as ext3 for the first time, ext3 initializes the \
> > journal. I suspect this initialization process might be interfering with \
> > something presto does. AFAIK, tune2fs on the other hand initializes the journal \
> > for you. 
> 
> Ah. Yes, I thought about some kind of conflict too. I think it makes sense to \
> recommend tune2fs instead of manual "dd" in the docs. 

 I will make the change in the howto. The dd method of creating ext3 journals is no \
longer recomended anyway.

> 
> 
> > Oops, I meant fuser -km...
> > 
> 
> Ok. What about startup? Any good way to run lento in daemon mode? I have a small \
> script which I usually use with "beta" programs, which writes all output in logs \
> and automatically restarts programs when they die, notifiying root via email.

 I have something simillar for our systems. Just make sure to kill lento before you \
close the journal.

> > > Also, I did some simple benchmarking when copying a big file (57Mb) my speed is \
> > > about 900 Kbyte/s. Network is 100Mbps. I do not plan to move such big files, \
> > > but maybe there are tunning options etc?
> > 
> > For now the best you can do is to bump up BLOCK_SIZE in Connection.pm to a larger \
> > value (and reinstall).  This is maximum the size of a chunk that may be \
> > transferred. I have used values upto 1MB for very large files, but I recomend \
> > around 250KB or so, esp. if you have many replicators. Please make sure that you \
> > have enough memory. For very large BLOCK_SIZEs you also need to increase the \
> > MAX_SILENCE timeout (just below the BLOCK_SIZE in Connection.pm), otherwise you \
> > will keep getting disconnected and re-transferring the files. Note that these \
> > values can be independantly set at each replicator and the server (i.e. they \
> > don't all have to be in sync). 
> 
> Actually the more I increased block size, the slower it became. On 64kb I got 6 \
> megabits, on 256 - 2 megabits. When I changed it to 4096, I got 7-8 megabits - same \
> as on 8096. On 1024, I get 4 megabits.

 Of course, on a LAN this makes sense. On WANs, which is what we use at Tacitus, the \
opposite is true.  I am surprised at the low troughput though, I was consistently \
able to get 2-3 MBps here on a 10 Mbps LAN.

 Is debugging output still on? That significantly lowers throughput. I

-Shirish



_______________________________________________
intermezzo-discuss mailing list
intermezzo-discuss@lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/intermezzo-discuss


[prev in list] [next in list] [prev in thread] [next in thread]