Re: 20GB IDE and Redhat 6.0

Jon Kay wrote:

> ALl right, but numbers like '8m per gig' have been bruited around a lot, and
> not without reason. Even 386m is alot more than 160m. Would you care to post
> a cachemgr.cgi run, maybe even a netstat -m run as well against that cache, so
> we can see something of what's going on?

The Squid process is "only" little more than 300MB. Of this at least
80MB (maybe more, possibly as much as 120MB) is from the 64bit
architecture of Alpha when compared to a 32 bit architecture. The other
200MB is for DNS server and resolvers, redirector, disk I/O buffers and
network buffers.

> I wonder if it's due to the high workload. Are request-related data structures
> and numbers of connections growing pathologically?

No. It is chewing along nicely. There are no abnormalities in the
balance of allocated data structures, except for the fact that
StoreEntry is quite a bit larger than on an Intel or other 32 bit
equivalent processor (104 bytes, compared to 56 on an Intel x86). Some
of the other structures are a also a little bit larger, but not that
much.

In round figures Squid on Alpha uses 90MB / 500K objects plus in in
transit data and some MB.

--
Henrik Nordstrom
Spare time Squid hacker