RE: Help!!

So, I'm going to have to ask.

If I'm building a new squid box, out of resources I have "just lying
around"....
        FreeBSD 4
        Compaq 3200 RAID
                2 x 24GB RAID (4x6 RAID 0, x2)
        Dual PentiumPRO 200
        512MB RAM

If I mount the RAID packs NOATIME, and with SOFTUPDATES enabled is there any
benefit to be gained by configuring squid itself with ASYNCIO?

Has anyone tried/tested this?

Thanks,
Phil P

> -----Original Message-----
> From: Joel Jaeggli [mailto:joelja@darkwing.uoregon.edu]
> Sent: Wednesday, July 05, 2000 7:27 AM
> To: Merton Campbell Crockett
> Cc: David Luyer; jubaco30; squid-users@ircache.net
> Subject: Re: Help!!
>
>
> On Wed, 5 Jul 2000, Merton Campbell Crockett wrote:
>
> > On Wed, 5 Jul 2000, David Luyer wrote:
> >
> > > > On Mon, 3 Jul 2000, jubaco30 wrote:
> > > > > I need to know urgently(An Administrative Decision)
> if squid takes
> > > > > advantage of SMP (Redhat 6.2 with 2 processors). The
> main questions is
> > > > > if the granularity of locks in I/O lead to similar
> performance(with
> > > > > smp, without smp)?.
> > > >
> > > > It is my understanding that Squid is implemented as a
> single process with
> > > > its own internal state machine/executive to manage its activity.
> > >
> > > Under Linux it multi-threads if you enable ASYNCIO, which
> means it scales
> > > much better over multiple separate disks (without RAID).
> > >
> > > > Under BSD/OS 4.x, Squid's apparent performce improved
> dramatically on
> > > > multi-processor systems. The observed performance
> improvement on the dual-
> > > > and quad-processor systems that we have is more a
> function of how well the
> > > > OS manages its resources and activities in an SMP environment.
> > >
> > > Under BSD/OS you're likely to not be using ASYNCIO, so
> you have limited
> > > gain from SMP for Squid, but there is still some gain there.
> > >
> > > > While we do have a few Linux systems, we tend to use
> BSD/OS where the
> > > > leather meets the road. This is probably more a
> function of two decades of
> > > > experience with the Berkeley Software Distribution from
> CSRG and BSDi.
> > >
> > > Well we use it too (as a result of policy, not of choice)
> but the number
> > > of bugs and lack of development tools is just damn
> annoying to people
> > > who are used to something as robust and well supported
> for development as
> > > Linux. Fortunately Squid doesn't hit many of the bugs
> (but OTOH it doesn't
> > > do ASYNCIO under BSD/OS which is a bit of a performance
> pain, but can be
> > > countered by just buying faster boxes than you would need
> to with Linux).
> >
> > I gather from your comments that ASYNCIO is similar in
> nature to the BSD/OS
> > file system's "soft update" feature. As I recall from a
> colleague's tests,
> > the performance improvement from using "soft updates"
> eliminated the need
> > to add a second processor to systems we were designing for
> our customers.
>
> softupdates on freebsd increases the performance of
> create/write/destroy
> operations by a factor of like 300 in my tests, that's the difference
> between being able to 30 create/write/destroy cycles on the
> same file per
> filseystem per second, and being able to do ~9000. I'm not
> real sure how
> much smp would help in the situation, but as far as I'm concerned ufs
> without softupdates is broken. A similar (but not identical)
> linux box I
> have can do about 11,000 with async io turned on (the
> default) and while
> I will say that ext2 is more brittle than ufs in the event of
> crash I'm
> noot sure I care with proxy cache filesystems.
>
> > Interestingly BSDi doesn't tout this feature all that much.
> I guess it
> > doesn't have the sex-appeal of the SMP cachet.
> >
> > Merton Campbell Crockett
> >
>
> --
> --------------------------------------------------------------
> ------------
> Joel Jaeggli
> joelja@darkwing.uoregon.edu
> Academic User Services
> consult@gladstone.uoregon.edu
> PGP Key Fingerprint: 1DE9 8FCA 51FB 4195 B42A 9C32 A30D 121E
> --------------------------------------------------------------
> ------------
> It is clear that the arm of criticism cannot replace the criticism of
> arms. Karl Marx -- Introduction to the critique of Hegel's
> Philosophy of
> the right, 1843.
>
>
>
Received on Wed Jul 05 2000 - 10:43:38 MDT