Re: [squid-users] Xeon / P3 / Dual decision

From: Ahsan Ali <ahsan@dont-contact.us>
Date: Thu, 8 Nov 2001 19:53:58 +0500

I checked out the reviews and it does seem like the perfect server - the
only thing I dont like is the maximum of 4 drives - as per marc's
suggestion, I was considerring 4 x 18GB (15K RPM) U160s but thats not a lot
of capacity... especially if there is a boot partition etc in there. I
suppose Fibre Channel storage is an option there.

Has anyone ever used NASs with Squid? How exactly does a NAS work? [Network
Area Storage] - Does it attach to a single machine and remain dedicated to
it, or can multiple hosts share it? If so, it must be slower as the storage
is probably done at a higher level than hardware... if someone has pointers
to reading material on this I'd greatly appreciate it!

Thanks all for all the input you guys have given - the Athlon MP does seem a
wonderful option.

-Ahsan Ali
----- Original Message -----
From: <sean.upton@uniontrib.com>
To: <squid-users@squid-cache.org>
Sent: Thursday, November 08, 2001 1:51 AM
Subject: RE: [squid-users] Xeon / P3 / Dual decision

> My company is also looking into an Athlon MP solution; the Appro 1124
> configured with internal RAID, Ultra160 drives, and lots of DDR memory
looks
> to be one possibility my company is looking into for a pair of 2
> high-traffic Squid accelerator boxes for next year. We rely heavily on a
> redirector, which is taxing, so high-system speed is going to be important
> to us. Right now we use Sun Enterprise servers (running Linux), which
> aren't exactly great in the CPU department, but are nice in the overall
> hardware design department, this isn't beneficial to us in our heavy use
of
> redirectors. My best educated guess is that Athlon MP systems will do
this
> fastest, based upon benchmarks I have seen. We use a python-based
> redirector, and the new Athlons really excel at running Python fast (not
to
> mention improve communication between Squid and the redirectors); also,
the
> advantage of using 266MHz DDR RAM is quite nice, I imagine, for
> frequently-used in-memory cached objects.
>
> There is, AFAICT, only one server-grade dual Athlon MB right now (Tyan
> thunder K7), which may be a drawback, but it has received a lot of
excellent
> review, though it does have fairly serious power and cooling requirements
> (the Appro box that my company is strongly considering, according to the
> AnandTech review I read, is designed to deal with this from the start
> http://www.anandtech.com/IT/showdoc.html?i=1514&p=3). The other
> disadvantage to the Tyan board is that it doesn't have 64-bit PCI support,
> at least in its current revision. Hopefully once I eventually get my
hands
> on these boxes, I'll be able to do a few informal benchmarks and
> stress-tests...
>
> I imagine, though, for raw latency, and not using helpers such as auth,
> redirectors, or dnsserver processes, your biggest issue is not CPU speed
or
> quantity, but system bus (FSB and peripheral) bandwidth, memory speed, and
> storage throughput. Still, the Athlon MP platform may excel at this better
> than any Xeon offering (AnandTech's database benchmarks might reflect its
> potential advantages in a small-read-io-intensive situation where the CPU
> speed is not that big of a deal -
> http://www.anandtech.com/chipsets/showdoc.html?i=1483&p=13).
>
> Also, dual CPU with respect to either the MP Xeon and Athlon MP systems
> should buy you at least some better overall system performance including
raw
> IO on a heavily saturated system. Gbit ethernet may be overkill though for
> 8Mb of IP traffic, though supposedly it often has a marginal latency
> improvement over 100Mb ethernet.
>
> Again, these stats are purely hearsay, and I haven't personally had a
chance
> to play around with this hardware, but I think its a smart guess that this
> is the way to go - at least the way I plan to go when I buy replacement
> servers for my outdated Sun equipment. If nothing else, from a
> performance/value standpoint, it makes good sense. I agree with Andrew:
for
> the price (I think about $3500-4000 each, loaded), you could likely get 2
> boxes like this for the price of a dual-Xeon box, though that might be
> overkill.
>
> Sean
>
> -----Original Message-----
> From: Andrew [mailto:tx3turbo4wd@yahoo.com]
> Sent: Tuesday, November 06, 2001 11:23 PM
> To: Ahsan Ali; squid-users@squid-cache.org
> Subject: Re: [squid-users] Xeon / P3 / Dual decision
>
>
> Seriously if you want performance you should be looking
> at a dual Athlon MP system. In most benchmarks these
> wipe the floor against dual Xeon systems.
>
> Or you could save yourself some $$ and just get a single
> Athlon XP 1800+ with 1.5gig DDR ram and then with the
> $$ you save you could prolly have a hot backup machine
> just using IDE drives.
>
> Linux and Reiser FS would be the OS and filesystem
> of choice.
>
>
> Cheers
>
> Andrew
> ----- Original Message -----
> From: "Ahsan Ali" <ahsan@khi.comsats.net.pk>
> To: <squid-users@squid-cache.org>
> Sent: Wednesday, November 07, 2001 4:46 PM
> Subject: [squid-users] Xeon / P3 / Dual decision
>
>
> > Hello guys,
> >
> > I'm building a Squid proxy to handle approximately 8Mbit/sec of
bandwidth.
> > I've decided to use the following:
> >
> > 3 x 36GB SCSI hard drives
> > 1 x 18GB (OS + logging etc - logging only when needed)
> > 2 GB RAM [may use 3]
> > GBIT Ethernet (may end up using Fast Ethernet only)
> >
> > I need to maximize bandwidth on this box without completely sacrificing
> > latency. I will not be running any content/url/regex checking for
blocking
> > sites on this proxy - so I think a dual processor system would be a
waste.
> >
> > But I have the following choices with respect to the processor(s):
> >
> > 1 x P3 - 1.26GHz w/ 256K cache
> > 2 x P3 - 1.26GHz w/ 256K cache
> > 1 x P3-Xeon-700MHz w/ 2MB cache
> > 2 x P3-Xeon-700MHz w/ 2MB cache
> >
> > Which one should I go for?
> >
> > Also, when its time to upgrade this box, shall I add more ram and disk
or
> > just add another identical proxy and peer it with this one?
> >
> > Thanks in advance!
> >
> > -Ahsan Ali
>
Received on Thu Nov 08 2001 - 08:44:44 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:03:59 MST