[OpenAFS] Re: WAN speed

jukka.tuominen@finndesign.fi jukka.tuominen@finndesign.fi
Fri, 23 Mar 2012 23:30:32 +0200 (EET)


> On Fri, 23 Mar 2012 01:02:10 +0200 (EET)
> jukka.tuominen@finndesign.fi wrote:
>
>> > 109M single file:
>> >
>> > SSH to AFS ~525KB/s
>> > AFS to SSH ~800KB/s
> [...]
>> I tried turning off encryption, but it didn't make a notable
>> difference with either file transfer test
>
> Okay, well that's obviously better but still much lower than the
> theoretical. There are a few more switches you can fiddle with:
>
>  - increase the chunksize by passing -chunksize to afsd (you have to
>    restart the client to change this). Try '-chunksize 20' for 1M, or
>    '-chunksize 23' for 8M, or something around there.

I'm propably not doing this correctly, but hopefully we're on to something:

1st trial
# afsd -chunksize 20
> afsd: Error calling AFSOP_CACHEINODE: not configured

2nd trial
# invoke-rc.d openafs-client stop
# afsd -chunksize 20
> afsd: Error -1 in basic initialization


How to do this correctly? Asking the basics, sorry

br, jukka


>
>  - try memcache instead of disk cache (I assume you're using disk
>    cache). Do this by passing -memcache to afsd, but this will use RAM
>    instead of local disk for the caching stuff, so you might need to
>    lower the cache size in 'cacheinfo'.
>
>  - try 'cache bypass' (with disk cache) which you can turn on with 'fs
>    bypassthresh'. But keep in mind cache bypass is still a new feature
>    and may not be entirely stable.
>
> If it still seems slow no matter what you try, what may be helpful is if
> you can provide the output of:
>
> rxdebug <client> 7001 -rxstat -noconn
> rxdebug <server> 7000 -rxstat -noconn
>
> So we could see the rate of resends and such for Rx. Provide that output
> both immediately before and immediately after you try a transfer.
>
> You can also try a benchmark using 'rxperf' to see how fast rx itself
> goes, which could help rule out what the bottleneck is during the
> transfer. If you want to try that but are not sure how to use it, let us
> know.
>
> --
> Andrew Deason
> adeason@sinenomine.net
>
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info
>