ted creedon wrote:

> General comment:
> 
> At the Workshop the MIT presenter stated that AFS worked fine until it was
> "placed under load".
> 
> In the process of copying files from a 1.2.11 box to a 1.3.87 box I ran into
> the same problem with Linux 2 Linux copying. That's how all this started.
> 
> tedc

David Tanner was referring to the SMB/CIFS server in the AFS Client
Service.   Under heavy load the SMB/CIFS server would become
unaccessible.   This appears to have been the result of the SMB/CIFS
error messages that were being returned when a timeout took place.
The SMB/CIFS client in Windows would disconnect from the AFS Client
Service and begin to report "Network Name Unavailable" errors to the
application even though the AFS Client Service was working just fine.

There are no problems with the AFS Cache Manager under heavy load on
Windows.

As for the "Network Name Unavailable" errors, I think I have corrected
this problem for the 1.3.8800 release.

Also, "load" in the case of the Windows Stress Test meant that the
SMB/CIFS client was transmitting almost double the number of outstanding
requests to the AFS Client Service than can be supported by the Windows
architecture.   There is a max limit of 63 simultaneous requests
and the stress test was generating over a 100 at a time.

Jeffrey Altman



Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to