> >> --- On Wed, 7/16/08, Adrian Chadd <adrian@xxxxxxxxxxx> wrote: >> >>> From: Adrian Chadd <adrian@xxxxxxxxxxx> >>> Subject: Re: Squid requirements >>> To: "Chris Robertson" <crobertson@xxxxxxx> >>> Cc: "Squid Users" <squid-users@xxxxxxxxxxxxxxx> >>> Date: Wednesday, July 16, 2008, 9:28 AM >>> What we're really missing is a bunch of "hardware >>> x, config y, testing >>> z, results a, b, c." TMF used to have some stuff up >>> for older hardware >>> but there's just nothing recent to use as a measuring >>> stick.. >>> >> >> The problem is that there's so much disparate technology out there. >> multi-core cpus, all kinds of different memory, all kinds of different >> disk >> technologies, different filesystems, different OS, different kernels, >> and on and >> on. It's hard to get useful measuring sticks. >> > > shoot me, but as ever faster is more expensive, so if you can't afford a > Lamborghini but like what it does then buy something else what comes close > and fits > your budget, hammer-speed and cheap does not exist, reasonable speed at > reasonable > cost does exist, hammer-speed at low-cost does not exist unless you jump > the cliff > what might result in sudden-death ... that is free and is fufufast > (sudden=>now) > >> I still think it's a useful pursuit. But I think that the reasons above >> make >> people less inclined to do it.ree and > > > to do what? caching? or proxying? or nothing? > while(my_input=0); (do='nothing'); > >> >> spec.org tries to level the field, if someone concocted a level field >> and made it >> easy for people to do, then we'd see more results. >> > > problem is most people look for easy=>lazy and lazy=>cheap but > unfortunatly that > equation does not work either Wrong. It works. Just not very fast. :-) I've had squid running on a 800MHz machine with 10GB HDD. 25% savings on the web bandwidth costs paid for the upgrade machine in short order. Now the savings are even better. > > as also do not exist any valuable hardware comparism since you need to do > it > yourself, means you need to look (clients, uplink, machine, > bandwidth_for_each, > disired_performance, budget) and finally look at your cache and at the end > it is > what_you_get_is_what_you_get_(for_your_money) ... so my friend, at the end > it does > not matter what they say to buy what you _CAN_ buy and get lucky with it > :) The measure of interest to an admin spec'ing up a Squid box are IMO: - highest req/sec vs cost. H/W config provides the cost scale. But whats missing is: what req/sec matches what H/W config for Squid? Amos
