[squid-users] Squid-Tuning: Can I know the how much latency the proxy adds?

In my network I have a squid proxy that reaches 50% request hit and 20-25% bit hit rate (a cyber cafe with 6 client machines). From this readings (obtained with "calamaris") I would supose my clients have a "enhanced browsing experience".. However, I am not sure if I'm right or not, cause I dunno how things seem to the end-user..

I've also seen that more about 50% of my request are for ".com", so I even tought of getting a "clustered proxy" where I would have a dedicated squid machine for caching the ".com" and the other one would cache the remaining things (I am using old machines that can only old 256 Ram max, so I cannot just add ram and disc tho the one I'm using to reach a value as I can have with 2 machines).

I would like to understand how having squid (or a clustered squid, where you would could have 2 machines "in-the-middle" in the worst scenario) affects the end-user, and I tought that the request latency would be what the end-user mainly sees.. (I may be wrong). Also, as I use a old machine (a P200 Mmx with 256 Mb ram) as Nat Router, Dns proxy/cache and Squid Cache (1+1 Gb in two ide drives), maybe the latency introduced is somewhat higher than it would be if I used a better machine?

Well, I would like to find out how better the connection would be if I would not have squid in between, so how could I know this? Lets supose the requests would not be cached, how do I know how worse is having squid in the middle instead of not using a proxy? I think I cannot have those values from squid log, can I?

There are some "tuning" details that I would like to figure out, for instance (everything I say is based on calamaris reports): I have "heap gdsf"/ LRU in my mem/disc removal policy, instead of "LRU/LRU". I've found out that like this my average hit replies are faster than before (more Kb/sec) however my "mem hit kb/sec" are lower, being almost the same as my "(disc) hit kb/sec". Before, I would clearly see a huge diference between "mem hit kb/sec" and "(disc) hit kb/sec"

I think I know why this happens: the "heap GDSF" keeps more (smaller) objects in memory, so I have more hits. But these smaller objects also have more overhead being returned to the client than the bigger objects would have, so I see a lower kb/sec using heap gdsf than using lru...
This is what confuses me: My clients "feel" the connection is faster now, or not? Should I care with this "mem hit rate" value or it's irrelevant when I see that "average hit rate" is better?

If it is relevant for anyone, I use squid 2.4 update 7, with "adzapper", on a ISDN 64K connection.

-- 
João Pedro Clemente  -  jpcl (at) rnl ist utl pt
" Why can't women put the toilet seat back up? "