Re: Squid Mem + HugeTLB

From: Kinkie <kinkie-dev@dont-contact.us>
Date: Mon, 14 Nov 2005 10:17:42 +0100

On Sat, 2005-11-12 at 16:55 +0100, Ronny T. Lampert wrote:
> Hi,
>
> I was thinking if it made sense to use HugeTLB for the squid memory/buffer
> subsystem and was about to hack it in.
>
> I don't know how big other people run their memory caches, but mine is sized
> around 256MB - which makes it > 65000 pagetable entries (at 4kb/page) which
> absolutely kills any TLB or caches when walking the tables on a miss
> (locality issues aside). It further is a stressing factor on the VM subsystem.
>
> With HugeTLB (normally has 4MB pages) there would be only 66 which fits most
> TLBs entirely.
>
> What do you developers think - worthy addition?

To me it makes sense: chunked mempools make for good memory locality.
You'll probably have to hook into that code to make the allocations
consistent with the huge pages.

Since you don't mention it: is this a 3.0 development?
In my opinion it is not a project suitable for a 2.X development.

        Kinkie
Received on Mon Nov 14 2005 - 02:17:57 MST

This archive was generated by hypermail pre-2.1.9 : Thu Dec 01 2005 - 12:00:15 MST