[PROXYPER] linux 308 proxy needs prune juice and All-Bran.
ard at scms.waikato.ac.nz
Thu Aug 12 10:05:33 EDT 1999
> I've got about 37k blocks in my out buffer that won't
> flush. The blocks were actually gathered by the
> Solaris 308 proxy and then the rc5 files transferred
> to the linux box for flushing.
I had the same situation, with 110,000 blocks gathered on a Solaris proxy
that jammed. Turning of duplicate checking helped enormously, by the
I transferred the blocks to a Linux machine, having some doubt about the
Solaris proxy, and flushed them from there.
> 08/11/99 15:08:55,rc564 r=0/21000, d=37324/3, 0.0 Mkeys/sec, tot=0
> 08/11/99 15:09:30,server: Successfully put back 15 unsent blocks.
> 08/11/99 15:10:02,rc564 r=0/21000, d=37324/3, 0.0 Mkeys/sec, tot=0
> 08/11/99 15:10:02,rc564 last serverconnection sent 63 and received 0
That last message is a bit misleading, since all sixty three 2^28 blocks
were returned in fifteen, ummm, blocks, for a net change of zero.
This is where my situation differed from yours. My Linux proxy would
always manage to send a fifteen or thirty blocks through to the server
before I lost the connection:
Got ack from server of rc564 654dbaa4:90000000, iter 1
So I tried a bunch of different central proxies, settled on one that
seemed the most helpful, and set a cron job to ALRM the proxy every two
minutes. After a night, the 110,000 blocks had been whittled down to a
few tens of thousands, which flushed out with a couple more ALRMs.
It seems the proxy has a problem sending blocks when it has a huge number
in the out buffer. The fewer in the out buffer, the more it can send
Andrew Donkin Technical Support
School of Computing and Maths, Waikato University, Hamilton, New Zealand
To unsubscribe, send 'unsubscribe proxyper' to majordomo at lists.distributed.net
More information about the proxyper