On Thu, Sep 03, 1998 at 06:34:58PM +1000, Andrew Hall wrote:
> Hello all,
>
> Need some help and I'm a little embarrased that I don't know this,
> but I've just upgraded one of my squid machines (linux 2.0.35 +
> squid 1.2b24p1) from a single CPU to dual PII300 and upgraded
> memory from 64MB to 1GB. I have recompiled kernel and squid for SMP
> enabled but:
>
> 1. How do I know if my machine is now using both processors?
> Does Squid make use of them? How much performance increase
> (if any) should I expect on what was a fairly heavily
> utilised machine?
Linux 2.0.x doesn't scale well when applications are IO bound (it
uses a global kernel lock, 2.1.x uses more finely grained locking),
but thats not so say you can't get some advantage from MP.
Squid is normally one large monolithic process. To get better
parallelism out of it, install libpthreads in rebuild the squid with
'./configure --enable-async-io'. (This helps greatly on single
processor machines too).
libc5 is a bit evil with libpthreads, you really need a glibc2 system
if your not prepared to hack about with this one.
> 2. Squid doesn't seem to find the rest of my memory (nor does
> the OS for that matter). I thought a recompile should have
> fixed this? How do I get my machine to find all my RAM?
Add mem=1024M to the append flags in /etc/lilo.conf, run lilo and
reboot.
Linux 2.0.x doesn't know how to find more than 64M of memory. (I
think Alan is looking at this for 2.0.36 though?)
-cw
Received on Thu Sep 03 1998 - 01:47:51 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:41:51 MST