Amos Jeffries wrote:
william wrote:
Chris Robertson wrote:
william wrote:
i would assigh 12GB memcache and four disks with diskd to each
instance.
Last I heard, diskd was not the recommended cache_dir algorithm
under Linux. If you are looking for maximum performance, Squid 2.7
with a combination of COSS and aufs is the current route to take.
On the other hand, the 3.1 branch of Squid is where the majority of
development effort is being focused.
I thought about diskd for some threading. But if COSS is the way to
go i will set that up.
I also think about squid 3.1 because of the nice features implemented.
3.1 does not provide COSS and is still a little bit slower than 2.7.
Amos
Amos
how little slower?
mean 5000 concurrent connections less or maybe a litle bit slower like
like 200 connections.
Or is it in the disk subsystem?
And what about stability and features like ntlm passthrough auth?
i tested this with squid 3.1 which works perfectly.
But 3.1 latest has no fix for bug 2155 and i get those once in two days.
I know this is fixed in the tree as stated in the bug report.
Any further recommendations for disk / os tuning?
With kind regards
William
