[squid-users] Question on throughput

[Jacques Kruger]

Thanks for the information.

I'll do some further testing and confirm that the CPU isn't the bottleneck in this case. The machine is a bit long in the tooth but with the faster connection this could easily be the issue.

As a general rule, am I correct in sticking with 2.7 for performance or is the current focus on squid3? 

Regards,

Jacques

-----Original Message-----
From: squid-users [mailto:squid-users-bounces at lists.squid-cache.org] On Behalf Of Amos Jeffries
Sent: 15 October 2014 11:31 AM
To: squid-users at lists.squid-cache.org
Subject: Re: [squid-users] Question on throughput

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 15/10/2014 9:41 p.m., Jacques Kruger wrote:
> Hi,
> 
> I've implemented my fair share of squid proxies over the past couple 
> of years and I've always been able to find a solution in the mail 
> archive, but this time around I'm stumped. This is the first time I've 
> used squid with a fast (in our context) internet connection, 
> specifically a 4G connection that the provider claims can run up to 
> 100Mbps. Claims aside, my real-world testing is not what I'm 
> expecting. I've used two squid instances, one on PFsence
> (2.7.9) and one on Windows (2.7Stable8) and compared the throughput to 
> a connection without squid and what I've found is, when testing with 
> www.speedtest.net<http://www.speedtest.net> the throughput is roughly 
> half with squid compared to a direct connection. I've left to 
> configuration pretty much default and have tried to tweak, both 
> without success.
> 
> What are the directives that have the most effect on throughput?

Not directives particularly, but on Windows the FD limit is fixed at an absolute 2048, whereas non-Windows can exceed that by a few hundred thousand or millions if needed.

It also depends on the NIC of the Squid  machine. If that Windows box is using a single NIC, then you will be maxing out the NIC capacity with traffic going over it twice (client->Squid and Squid->Internet).

Then also the CPU gets a say. If the Squid is not doing much and the CPU is very fast, Squid can end up running a lot of work cycles transferring just one or a few bytes. Which impacts the TCP overheads by decreasing bytes-per-packet.

It also depends on the protocol being used by the tests. In the modern web there are HTTP/1.0, HTTP/1.1, WebSockets, HTTP/2, and SPDY capable of being used for the test itself - the latter two of those involve bandwidth comoression that can greatly enhance throughput. But only
HTTP/1.0 through default Squid-2.7 (HTTP/1.1 with tweaking).


PS. I would be interested to see what your results are with the
squid-3.3.3 now available for Windows via Cygwin.

Amos
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (MingW32)

iQEcBAEBAgAGBQJUPj7eAAoJELJo5wb/XPRj0WgH/iBxr/0CK+MMnL5PB/ZyANqx
uxazgxCsp+EDyxl2MR+iccsaerN/gdTM7Rzms1/PBlExk2yNQb8/6EUC8XQlJs8m
EEDI0lmd13jeVIWmXP2vYy1vQATuS7OMI0bvkfCDSI/bYVaHKMKCtssCPdQRKxul
akGgY7LhX7lLlkzgzrnrU8HUNN/PjFI7tg3y0xKihwEaDckYHcqg260cHJdjeI6C
2kZfFeJFMV0D0Y7N1HO38OKkJpHBIJoWDWWWlmI8j01UpsBSuxRqw51r1372jkne
Ka5m9vlQfGXWvI5Zg7uL3CxlBRTe6oe7T0n3dmHGy4KnvDbT09MXYMkS3ShTfzU=
=7UJb
-----END PGP SIGNATURE-----
_______________________________________________
squid-users mailing list
squid-users at lists.squid-cache.org
http://lists.squid-cache.org/listinfo/squid-users