Can I run multiple overchan on CNFS/OVDB safely?

[Heath Kehoe]

>
>Is there any reason not to run multiple overchan processes?
>

In theory, you should be able to run multiple overchans.  I've not
tried it.  You may not see any performance improvement, as the overchans
may end up getting in each others way (one getting blocked on another's
lock).


>I am trying to track down a performance problem on my 2.4 installation.
>

Have you looked at http://www.avalon.net/~hakehoe/inn/tipsandtricks.html ?


>
>My feed machine/aggregator is sucking in about 200GB/day, no problem.
>
>My reader machine gets it's only feed from this aggregator, and nothing
>I can do seems to get it to suck in data fast enough.
>
>There are long pauses on the reader machine with disk I/O on the overview
>drives is extremely high, and if I watch netstat -w  1 output, I receive
>about 2-3 MBytes/second for about 20 seconds.  THen it slows waaaay down to only a few K
>per second for 10-15 seconds.  AT the same time systat -iostat  is shoinw 
>intense pounding on the overview disks.
>
>THen it picks back up again.

That could be the checkpoints slowing you down.  You can make checkpoints
more frequent, which should make them complete more quickly and thus
even out the load.  To do that you'll have to edit frontends/ovdb_monitor.c
(eventually, I'll add more config parms in ovdb.conf for this).
Find the txn_checkpoint() calls.  In the while loop you should see a
sleep(30) around line 195.  Change the 30 to 5.


>
>I have tried splitting large and small articles, splitting off
>control messages, I'm just about out of brilliant ideas.
>

Are you using ovgrouppat: to not store control.cancel?


>SO I want to see if it's overview holding up the process, and wondering
>if it's legal/supported.
>

I'll have some time this week to work on it, so I'll make ovdb_monitor
more configurable via ovdb.conf.

-heath