Matus UHLAR - fantomas wrote:
On 02.08.09 01:59, Waitman Gobble wrote:
Matus UHLAR - fantomas wrote:
I hope you know what are you doing by configuring that big cache_mem...
http://wiki.squid-cache.org/SquidFaq/SquidMemory
hmmm, i've read that. i don't by the way,
yes, it seems soo..
yes, it's been several years since i've worked with squid. i'm not
worried about it though, i'll catch up.
however, nothing bad has yet happened, mem utilization seems ok, cache
hits are up. but it's only been a few days.
HTTP: 1166366 Requests, 710613 Hits ( 61%)
mem_total: = 4294967296 ( 4096MB) [100%] Logically total memory
so you have 4GiB of RAM and confiured 3GB for squid's mem_cache?
when that get filled up, you either start swapping or get ouf ot RAM and
your squid will crash...
i've kept an eye on it, and i agree that it may start paging and/or
crash. but so far, so good.
the most disappointing thing is the CPU load on the cache server, it
hovers around zero (0). once i noticed it was at .2 (20%) and thought
something was about to happen, but it dropped down to zero again pretty
quick. it's rather like having a server that doesn't actually do
anything, and you want to load it up with some email servers or database
or something just so it feels better.
you have low CPU usage and don't like it? Apparently your clients aren't
using the cache, or there is not that big traffic so the cache would get
loaded. Or there is another problem and your CPU spends its time waiting for
i/o...
i've built hundreds of servers (but not so many running squid) over the
past 12 years, and when it's running 0 load that actually feels like
waste to me. but at the moment i'm looking at the end result, which is
good. it's not getting so much traffic, just under 1 million page views
a month according to g analytics. i'm considering setting it in front of
some high traffic sites, but only after a while, maybe a month, when i
feel like it's not going to bomb and cause problems. so at the moment,
it's a live experiment. the biggest problem i've struggled with so far
recently is a big problem with the TSO implementation in freebsd 7 (i'm
running 7.2 on machines at the moment), which causes some clients to
pull data (pages, images) at a slow drag, like they're on a 9600 baud
dialup. and it was a tough one to track down, because it seems to affect
a small percentage of clients (guessing maybe 15-20% worldwide - enough
to worry about - but it took me a good long while to actually find a
machine 'in the wild' that behaved this way) - and literally two
machines side by side on the same remote network / LAN would behave
differently, one super fast and the other doggy slow - it has something
to do with the nic and settings on the client. i initially believed this
problem to be isolated to the fx_ drivers {there's an issue in the bug
db} but recently I've had complaints regarding a server with nics using
em drivers. So I'm at the point of *always* killing TSO
waitman
