On 10/03/2010 05:56 PM, Harald Barth wrote:
I digged into the /var/log/messages and found something:
Based on this I changed the client settings for the memcache. I went from
"AUTOMATIC" to "65536" (64MB) and that worked. Going up to 96MB worked,
128MB crashed the machine again. Tried 97MB and it crashed as well. 96MB
seems to be the limit.
There is a limit how much kernel memory you can allocate (that is what
memcache does) on 32bit Linux. That number is _much_ bigger on 64bit
Linux.
I think it should refuse to start instead of crashing. Another route to take
would be "disc cache" in tmpfs, but I don't know if that works nowadays.
It does. I remember trying that around 1.4.11, but never in a 32bit
environment (I don't see why that would be a problem, though).
Note that if the motivation for memcache is faster bulk transfers (which
you gain), disk cache in a ramdisk won't give you the same throughput.
But I agree, this approach could conceivably be much more stable in some
environments.
Regards,
Felix
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
