Re: [squid-users] Ideal cache placement (was Re: Why Squid is great (was: fourth cache off??))

Jon Kay wrote:
> > To demonstrate why, let's push this to its natural conclusion, where
> > each users has his/her own private cache. In this scenario, it's hard
> > to see how the caches are of much benefit, because of the lack of
> > sharing.
>
> I remember reading about this debating technique when I was 14.
> "Reductio ad absurdem." Let me try it on your argument.

If you want to debate, at least get my argument right. I
specifically said

# My gut feeling is that caches near real choke points make sense,
# and that caches near similar user populations also potentially
# makes sense. As for ideal placement, that probably depends more
# on topology than any blanket assertion can easily cover.

> OK, so if we should centralize everything, then let's have One Big
> Fast Cache in the center to serve all requests.

If there aren't any choke points between the users and that central
server, what's wrong with that argument?

> > In this scenario, it's hard to see how the caches are of much
> > benefit, because of the lack of sharing.
>
> ...unless they are running hint cache or Cache Digests or ICP or even
> old-fashioned simple hierarchies. In that case, there will be sharing.

Again, I already addressed this:

# Adding new communications
# into the mix may not be a clear win, especially if the extra
# communication doesn't scale well with the number of caches.

and

# The drawback to hierarchical caches
# is the additional latency involved in the hierarchy - much worse
# than router hops or line losses.

> Plus, I didn't suggest one box per user, but one box per internal group.
> In the real world, people setting up caches are able to make reasonable
> decisions about these things.

But again, why the arbitrary decision to aim for internal groups?
If the group actually generates enough traffic to make their
link to the internal backbone a choke point, then that's one thing.
Otherwise, it's just more administration for little gain.

Even from a price-performance standpoint, the approach of
arbitrarily peppering internal groups doesn't make sense. The
cheapest commercial cache at the 4th cacheoff was 130 reqs/sec
for $2500. The next one up handled 800 reqs/sec for $3250. If
I'm a system administrator at a company, I'd probably buy one
or two of the 800 req/sec boxes rather than 6 or 12 of the
130 req/sec units. LAN bandwidth is pretty cheap and not nearly
as lossy as you suggest.

Once you start going out to WANs, then it's a different issue,
and that's why I mentioned the choke points.

-Vivek
Received on Thu Dec 20 2001 - 12:32:25 MST