>> That's pretty much all that's needed. The 3 parent proxies are all
>> configured as siblings to one another.
>
> When you do cicular relations (including sibling) it is recommended to
> also use cache_peer_access to rule out loops. Deny forwarding to a
> peer if the request was received from one of your peers.
>
> I.e. limit the request routing between your caches to at most one hop.
Oh yes, I forgot to mention this. There're some ACLs and rules to
prevent this (which I picked up from some messages a while back, as well
as the book by D. Wessels).
>> The load distribution reported by MRTG seems to correspond with how
>> we wanted it. The proxies handle roughly 2000 requests/min on the
>> average (I wonder if its possible to increase that, though, as
>> we've peaked to 3000 once). The cache/hit ratio is pretty good as
>> well, around 25% on off-peak hours to as high as 80% on peak hours.
>
> 2000/min is a fairly low number.. (little more than 30/second) you
> most certainly can get a fair bit more out of your boxes unless you
> are running on really old scrap hardware found on the junkyard..
Thanks. The front proxy could be the bottle neck, but I doubt it. It
could also be that the surfing habits of users of the proxies. Either
way, I might try to get some benchmarking tools to see how much the
setup can really handle. Any suggestions on how to improve it? The
network setup is pretty good, with a rather fat pipe at that.
I was worried about this because there are times when several boxes
actually have to wait before one of the proxy handles them, and this
happens with different platform clients.
--------------------------------------
Gino LV. Ledesma
Ateneo Cervini-Eliazo Networks (ACENT)
email : ginoledesma@mac.com
web : http://cersa.admu.edu.ph/
phone : (63)(2) 426-6001 ext. 5925/5904
Received on Mon Jun 03 2002 - 22:50:10 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:08:25 MST