[OpenAFS] OpenAFS speed over the net

[ted creedon]

Nonetheless it is possible to empirically adjust parameters to improve
performance.

To the point of crashing. Some guidance is needed. 

Initial Windows client 1.3.87 crash tests with default settings are
acceptable. The same cannot be said for Linux.

I need to set up Tripwire to verify the round-trip data.

See: ftp://creedon.biz/pub/afs_stress_tests/run1 (and 0) for details on
crashing Linux. The test script is included.

Close but no cigar.

tedc

-----Original Message-----
From: Jeffrey Altman [mailto:jaltman@columbia.edu] 
Sent: Saturday, August 13, 2005 5:16 PM
To: ted creedon
Cc: OpenAFS-info@openafs.org
Subject: Re: [OpenAFS] OpenAFS speed over the net

With or without encryption?

Increasing the number of threads will not necessarily increase
performance.  It may very well slow it down.   There is a careful
balance that must be kept between the SMB/CIFS and RX thread pools.

Jeffrey Altman


ted creedon wrote:

> In previous tests with 1.3.84 Windws client:
> a 2 CPU 2.8GHz P4 client running at 80% on both processors (on the 
> same die, not a dual CPU box) loaded a dedicated 100BT net to about 
> 35% transferring files from a 1.2.11 server to the 1.3.84 Win client local
drive.
> 
> I did increase the cleint thread and process settings..
> tedc
> 
> -----Original Message-----
> From: openafs-info-admin@openafs.org 
> [mailto:openafs-info-admin@openafs.org]
> On Behalf Of Jeffrey Altman
> Sent: Wednesday, August 10, 2005 4:39 PM
> To: Frank Burkhardt
> Cc: OpenAFS-info@openafs.org
> Subject: Re: [OpenAFS] OpenAFS speed over the net
> 
> Frank Burkhardt wrote:
> 
> 
>>Hi,
>>
>>On Mon, Aug 08, 2005 at 05:02:53AM +0300, Jeffrey Altman wrote:
>>
>>
>>>The Windows client defaults to using encryption while the
>>>Linux client does not.   That is the most likely cause of the
>>>difference.
>>
>>
>>AFAIK Lars uses Debian and the Debian OpenAFS init-script runs 'fs 
>>setcrypt on' by default after starting the client.
>>
>>Regards,
>>
>>Frank
> 
> 
> Then I would look at the network configuration of the Windows
> box.   Using two very different hardware configurations on 100mbit/sec
> full duplex lans I am able to fully saturate the network at below
> 30% cpu utilization.   So I am fairly sure the issue is not entirely
> related to the windows client.
> 
> Jeffrey Altman
> 
> 
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info