Good day,
We are here approaching situation that more than 50% of all Web traffic
is porno ;( filling all available resources (disks, links, minds). As a
commercial ISP we cannot block it altogether, but I was wondering if it
is possible to arrange proxy-cache in such a way that business traffic
gets absolute priority over XXX stuff, and this lead me some thoughts.
For now we did a setup where primary cache has a parent with slightly
modified squid which inserts configurable stallDelay after every packet
for every session. Primary cache goes via this parent only if a special
xxx regex acl matches. In effect, xxx sites are accessible, but are very
slow and thus (hopefully) consume less bandwidth than normal traffic.
Also, we hope that this will reduce interest in those sites somewhat.
(Note that we enforce proxy-cache for international links, so have a
chance to play around with such things within cache setup)
As a quick&dirty hack it has its problems, waste of open files, for eg.
Also, while stall time ticks, incoming tcp windows fill up and next read
will give upto 16KB in a row (more than average object size)
For now, I've set stallDelay to 11 secs, ie. every miss stream is
serviced no more frequently than every 11 secs, giving approx. 1,6KB/sec
limit per each session. Primary cache hits are fast.
What I'd like to achieve is some kind of priority-queueing mechanism
on the primary cache itself.
Ideally, configration file using acl-s would send every URL request
into appropriate queue and these queues be serviced in some preconfigured
manner, upto limiting bits/sec for each queue. There might be fixed (8-256)
or variable number of queues and these be serviced in order of priority,
fulfilling each queue's requirements. Each queue could have properties
like max bytes to proxy in a go or in a second, per queue or per open
session. Sessions could be put into different queues based on URL, client,
sourceip or browser - whatever, all acl types included. Of cource, ICP
and disk io would go to priority 0 queue, thus reducing latency where
needed.
As this is actually more general resource management and is partly in
ToDo list, I wanted to know if anyone has been working in this direction
and what are general thoughts regarding this matter.
regards,
-------------------------------------------------------------------
Andres Kroonmaa Telefon: 6308 909
Network administrator
E-mail: andre@ml.ee Phone: (+372) 6308 909
Organization: MicroLink Online
EE0001, Estonia, Tallinn, Sakala 19 Fax: (+372) 6308 901
-------------------------------------------------------------------
Received on Tue Jul 29 2003 - 13:15:41 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:18 MST