>> Not being experianced with cache server demands, would this
>> be a good spec for a heavily used server to sit on a Global Network?
>>
>> MicroWay DEC Alpha "Screamer" Mboard
>> 500Mhz DEC Alpha CPU
>> 256Mb Ram
>> Adaptech 3940 UW Dual Channel SCSI Controller
>> 4 4.5Gb HDD's (One for /, logs, etc and 3 for cache spool)
>> Drives run at 5,400 RPM
>
>Talk about overkill! All that dedicated to cache?? What type of network?
>How many nodes simutaniosly served?
Not really, that's approximately what we're running (not as fast an
Alpha), but essentially identical memory and disk
configurations. We're serving around 2500 users, and we ship about 1Gb
per day (200 000 transactions). That consumes around half our
AlphaServer 1000A's 333Mhz CPU, although I'll admit that this box is
not optimally configured for squid.
Leigh asked about specs for a heavily used machine on a global
network, and I expect the server described above would probably be
appropriate (caveat emptor: without knowing the expected load, this is
a little speculative). I recommend *not* using Digital's advanced
file system on the cache spool, several factors in our recent analysis
are pointing at that as a bottleneck.
However, I have also come to the conclusion that the list seems to
recommend: several small-to-midrange caches are more efficient than
one huge cache. Even if you're starting with one cache, it's worth
assuming that you will subsequently move to several; make use of
proxy.pac files and have a think about your network topology. I am
currently looking to improve the performance of my cache by splitting
it across multiple boxes. I have asked my systems manager to provide
me with several servers that will run a kernel, cron, squid and
nothing else :-)
Jules Anderson
Received on Mon Jul 07 1997 - 15:04:59 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:35:41 MST