Re: large proxies

On Mon, Feb 21, 2000 at 10:51:38PM +0000, Miquel van Smoorenburg wrote:
> About 70 GB of disk. This is a dual Pentium III @ 450 Mhz,
> with 512 MB or RAM. Running squid-2.2STABLE5 on a Linux 2.2.15pre5
> kernel. The dual CPU is only useful if you have an OS with kernel-
> level threading and compile squid with async-io, otherwise squid
> would only use one CPU.

Unless you run multiple squid processes on different IP addresses,
though I guess that's a grossly inefficient way to do things given the
option of kernel threads and async-io :-)

> You don't really save a lot of bandwidth with a cache anymore, a lot
> of content is dynamic nowadays. But it does help for slow far-away
> sites that you visit reasonably often.

I think this trend is quite likely to reverse when broadband
connections take off worldwide. Some of the ad-serving companies
are getting smarter and now return a Location: to a cacheable
object. I'm sure some of the pointlessly-uncacheable dynamic
content will become cacheable. ColdFusion, for example, provided
an easier way to allow this quite recently (v4.5, not very much
used yet since it's still new and obviously it will take time
for many developers to trust it enough to upgrade production
servers), CF is responsible for a lot of the dynamic content on
the web. People are starting to become more clueful to the
benefits of caching and methods of implementation, languages
and tools are starting to provide easier ways to allow it,
I think it's just a question of time. And enough people using
high bandwidth connections and caches to make the bandwidth
saving at the webserver sufficient to justify the effort.
Received on Mon Feb 21 2000 - 16:36:10 MST