Brian wrote about "--enable-async-io"
> i didn't realize that squid would set it be default. the documentation is
> a little thing about that.
An older post by Henrik says it will set it to 16 by default.
> parse runs no problems. --enable-storeio=aufs is automatically set by
> --enable-async-io
Just checked with "configure -help" and you are right - guess I forgot...
> hmm, then i will leave it out, since all i need to do is pass SSL.
Well check the archives, but we don't have it and lots of people here do
online banking, etc.
> as of right now this is impossible, and not really needed. this error
seems
> to be harmless, people can still use the cache just fine. once we turn
off
> netscape and put this in its place, i can dedicate all three 9G disks to
> squid. i will also put VxFS on those disks since UFS on solaris is a
known
> issue with AUFS.
This is news to me. What makes you say that? I have my cache on an Ultra
60 running 2.8 and mount the cache_dir's with the mount options
"noatime,logging" but that is it. Still I've heard VxFS is a high
performance filesystem so I assume it runs in the kernel, not another layer
of indirection on top of the UFS base?
As to sizing it once you are ready use the FAQ
http://www.squid-cache.org/Doc/FAQ/FAQ-8.html#ss8.11 to calculate how much
RAM would be needed. I recall recently reading here that you shouldn't use
more than 80% of any disk, so assume you define 7GB per disk = 21GB X 10MB
per GB of disk = 210MB RAM + 20MB for safety = 230MB. FAQ says you should
have double this in physical RAM so 460MB. I hope I did this correctly -
reread the FAQ cited above to confirm. Also you may or may not want to bump
up your cache_mem. I went overboard and then tuned it down to 64MB.
My own Sun 2.8 server has 14GB of disk X 10MB per GB of disk = 140 + 20 =
160 X twice that for physical = 320MB + 64MB for my cache_dir. For what it
is worth, total RAM usage for my squid server hovers between 350 and 410MB
of RAM (out of 2GB available). As soon as we have money I'd like to buy
more external disks for more caches and then the extra RAM will get used.
One other thought: originally I had compiled squid as a 64bit app. Not
only was I told that it had no practical benefit for squid but I noticed
that given the larger address size, more RAM was used and some other
problems cropped up. So not sure if your server is in 64 or 32 bit mode,
but I'd stick to 32bit since it won't help to go 64 and seemed to increase
RAM usage (YMMV).
hth
adam
Received on Thu Aug 07 2003 - 12:31:05 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:18:46 MST