Re: [squid-users] load tpoxy wccp on multiple interfaces by smp ?

On 3/11/2013 5:22 p.m., Dr.x wrote:
> hi ,
>
> its just an updating idea ,
> we have 6000 users and we have 96 G ram and 24 CPU cores and DELR720
> hardware ,
> actually i want to use smp and want to handle them by squid
> Q1-from the user experience who tried squid smp , can my hardware handle the
> 6000 users ????

No. It can handle some amount of requests/sec and traffic/sec. But
"users" is not related to proxy capacity.

6000 users doing 1 req/day, even the footstool under my desk can handle
that load.
6000 users doing ~150 req/sec each concurrently, you need a monster
amount of CPU to handle that load.

Squid-3.3 can handle something like 2k - 20k requests per second on
average-cost modern hardware in a single worker. That is somewhere
around 50 - 150Mbps of HTTP traffic if you look at things in Mbps. The
numbers *will* vary greatly depending on many factors in your users
traffic profile.

24 cores should be enough to handle it, though be careful that it is 24
*physcal* cores. Ignore any hyper-threaded / virtual cores.

> ===========================================================================
> Q2-can SMP let me use two tproxy on 2 interface and share cores cahcing on
> the two interfaces ???
>
> i mean my server will have eth1 , eth2 connected to router
>
> eth1 is x.x.x.x
> eth2 is y.y.y.y
>
> squid will be listening yo tproxy x.x.x.x:1111
> and also will be listening to tproxy y.y.y.y:2222
>
> and each interface will have wccp service number .===> mean that many wccp
> services will be working
>
> agian ,
> i want to do that , because the traffic on 1 interface cant handle more than
> 1 G traffic "" " my router cant handle more than 1 G """,
> so i need to use 2 interfaces so that make network load distribution
>
>
> can squid smp handle what i want ????

Maybe. High confidence without certainty.

> without bugs ?

Er, Um.

Amos
Received on Sun Nov 03 2013 - 09:21:57 MST