Re: [squid-users] High load server Disk problem

Hi,

2010/8/19 Amos Jeffries <squid3_at_treenet.co.nz>:
> Your COSS dirs are already sized at nearly 64GB each (65520 MB). With
> objects up to 1MB stored there. That holds most Windows updates, which are
> usually only a few hundred KB each.
> I'm not sure what your slice size is, but 15 of them are stored in RAM at
> any given time. You may want to increase that membuf= parameter a bit, or
> reduce the individual COSS dir size (requires a COSS dir erase and rebuild).

I'm trying to increase the membuf= parameter on COSS, but I had to
reduce the cache_dir (from 65520) size in order to do it:

cache_dir coss /cache/coss1 60520 max-size=1048575
max-stripe-waste=32768 block-size=4096 membufs=100
cache_dir coss /cache/coss2 60520 max-size=1048575
max-stripe-waste=32768 block-size=4096 membufs=100
cache_dir coss /cache/coss3 60520 max-size=1048575
max-stripe-waste=32768 block-size=4096 membufs=100

Do you guys have any idea if changing from 15 to 100 is enough? Is
there a way I can test this?

As far as the slice size goes, I don't know either :)

I compiled squid with default COSS options, so I guess it's the 1MB
default stripe size. Is this it?

One last thing: Where there's a documentation commenting on the
real-life benefits on using --enable-coss-aio-ops? Is there any?

I could help running some tests and create some documentation for the
COSS wiki page, if you think it'll help other people...I just need
help with some pointers on how to go about testing it right.

Thanks,

- Robert
Received on Mon Aug 23 2010 - 16:11:57 MDT