RE: [squid-users] 50 requests per second

From: Joel Jaeggli <joelja@dont-contact.us>
Date: Mon, 24 Jun 2002 05:14:14 -0700 (PDT)

On Mon, 24 Jun 2002, Robert Adkins wrote:

> Aaron,
>
> First, I must state that I am far from an expert with Cisco products.
> However, I believe that it is safe to say that for the most part, Cisco
> routers have no moving parts, like Hard drives in a computer system. Due
> to this fact alone, they are considerably less likely to fail, in
> comparison with a server.

actually a large cisco (like a 6509) has something like a dezon large fans
in two or three fan tray's... the probability of mechanical failure is
non-zero and the potential consequences are large...

then you have to figure in the fact that your biggest source of router
outages (80%) is human error followed by os bugs and then hardware... all
those are good reasons to have redundant router infrastructure...

joelja

> Sure, there are people that have servers running 24/7 for ten years...
> Blah, blah, blah... There is still a likely chance that a hard drive will
> crash.
>
> Having a server with moving parts as a single point of failure is quite
> a risky venture. Due to the configuration possibilities of a server it
> could take 8 or more hours just to bring it back to the state it was
> before the failure. Of course, some or most of those services could be
> restored far sooner then 8 hours.
>
> Again, I am unfamiliar with Cisco routers, but I would imagine that if
> you have a back-up of the configuration and a known-good spare unit, the
> Cisco router configuration could probably be restored within a
> half-hour's time.
>
> Personally, I would rather rely on nearly anyone's hardware pure router
> as a single point of failure instead of a machine with a few spinning
> magnetic discs.
>
> Regards,
> Robert Adkins
> IT Manager/Buyer
> IMPEL Industries, Inc.
>
> -----Original Message-----
> From: Aaron Seelye [mailto:AaronS@et-n-m.com]
> Sent: Friday, June 21, 2002 2:58 PM
> To: mailinglistsquid-users@squid-cache.org; Mark.H.Price@AOC.STATE.NC.US;
> Robert Adkins
> Cc: squid-users@squid-cache.org
> Subject: RE: [squid-users] 50 requests per second?
>
>
>
> So your single point of failure is a Cisco 6509? :)
>
> Aaron
>
> > -----Original Message-----
> > From: Mark.H.Price@AOC.STATE.NC.US
> > [mailto:Mark.H.Price@AOC.STATE.NC.US]
> > Sent: Friday, June 21, 2002 10:18 AM
> > To: joe@swelltech.com
> > Cc: squid-users@squid-cache.org
> > Subject: Re: [squid-users] 50 requests per second?
> >
> >
> >
> > Yes, while it may not be necessary to use 2 load balanced boxes
> > for serving that many, it would be frowned upon in my environment
> > to have a single point of failure. Our network is focusing on a Cisco
> > 6509 we have that will be doing the load-balancing.
> >
> > Mark
> > ------------------( Forwarded letter 1 follows )---------------------
> > Date: Fri, 21 Jun 2002 10:44:32 -0500
> > To: Mark.H.Price
> > Cc: squid-users@squid-cache.org
> > From: joe@swelltech.com
> > Sender: squid-users-return-1@gcs.alias
> > Subject: Re: [squid-users] 50 requests per second?
> >
> > Load balancing isn't strictly necessary. We have a single box serving
> > 200+ reqs/sec at peak periods (dialup ISP workload, so
> > different than a
> > LAN, but still should make a huge difference in peak request rate).
> >
> > It just takes a big box, ReiserFS, and Squid with aufs filesystems.
> >
> > Mark.H.Price@AOC.STATE.NC.US wrote:
> > > I have successfully tested squid with transparent caching, and it is
> > > working well for about 100 users. I am going to be
> > deploying squid for
> > > about 5000 users. I estimate that the proxy will need to
> > easily handle
> > > 50 requests per second, and perhaps up to 80 or 100 during
> > peak usage
> > > times.
> > >
> > > Does anyone have experience with a cache server serving with such
> > > a large user base?
> > >
> > > I am undecided whether I will be using a tree or mesh, but
> > load balancing
> > > will be necessary. The platform of choice is RedHat Linux.
> > >
> > > thanks
> > >
> > > Mark
> >
> > --
> > Joe Cooper <joe@swelltech.com>
> > Web caching appliances and support.
> > http://www.swelltech.com
> >
>

-- 
-------------------------------------------------------------------------- 
Joel Jaeggli	      Academic User Services   joelja@darkwing.uoregon.edu    
--    PGP Key Fingerprint: 1DE9 8FCA 51FB 4195 B42A 9C32 A30D 121E      --
  In Dr. Johnson's famous dictionary patriotism is defined as the last
  resort of the scoundrel.  With all due respect to an enlightened but
  inferior lexicographer I beg to submit that it is the first.
	   	            -- Ambrose Bierce, "The Devil's Dictionary"
Received on Mon Jun 24 2002 - 06:20:59 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:08:47 MST