On 26/02/2012 6:59 a.m., Ben wrote:
Hi Amos,
On 22.02.2012 03:15, Steve Tatlow wrote:
Hi,
We are running squid as a transparent proxy, with dansguardian doing
the
content filtering. All traffic will be coming from localhost and no
authentication is required. Can someone tell me how I ensure there are
enough squid processes to support a large number of users (maybe 250
concurrent users)
None of us can tell you specific numbers. It is dependent on your
hardware and client traffic.
The thing to be aware of is that measuring in users is meaningless.
One user can flood the proxy, or some thousands could leave it idle
waiting for more work. Capacities are reliably measured only in
requests per second.
To get the details you seek measure and get some idea of how many
requests per second those users make at peak times, and how many the
whole structure is capable of handling.
Each Squid series has a theoretical limit which is hardware
dependant (3.1 can do about 800 req/sec on a dual core 2.2GHz CPU
etc). The configuration specifics you create and type of requests the
clients will reduce the capacity limit from there.
You mean to say that single squid instance can handle 800 req/sec on a
dual core 2.2 GHz CPU ? Can you elaborate it in details means how many
hdd have you used and is there any specific configuration do you want
to highlight....
We have a test machine which can reach that. Nothing special on the
hardware, and in active use running several other services. But the test
is a bit artificial. So I think overall its a reasonable sort of result.
Your mileage *will* vary.
As i tested single squid instance with 400-450 req / sec and it is
performing fine.Currently i deployed squid with 175 Mbps bandwidth
load.Now we plan to use it for 400 Mbps so it suppose be 800 or 900
http req / sec , Does single squid process handle such heavy load or ?
The fact you got past 50Mbps easily at ~400 req/sec tells me your
traffiic might be a bit unusual. On the ISP scenario I'm used to
estimating with most of the reports have needed two Squid to get over
100Mbps. Good news for you, bad news for forcasting the limits.
And what kind of h/w specification you suggest for such kind of load ?
At this point you have a Squid already to use as baseline. So you can
look at the resource usage CPU, memory, Disk I/O etc and guess (yes
guess) how much more load it can take before any one of those is maxed out.
Amos
