Datasheet

“main” (Installation and Administration) — 2004/6/25 — 13:29 — page 602 — #628

The question remains as to how long all the other objects stored in the

cache should stay there. To determine this, all objects in the cache are as-

signed one of various possible states. Web and proxy servers ﬁnd out the

status of an object by adding headers to these objects, such as “Last mod-

iﬁed” or “Expires” and the corresponding date. Other headers specifying

that objects must not be cached are used as well.

Objects in the cache are normally replaced, due to a lack of free hard disk

space, using algorithms such as LRU (last recently used). Basically this

means that the proxy expunges the objects that have not been requested

for the longest time.

25.3.3 System Requirements

The most important thing is to determine the maximum load the system

must bear. It is, therefore, important to pay more attention to the load

peaks, because these might be more than four times the day’s average.

When in doubt, it would be better to overestimate the system’s require-

ments, because having Squid working close to the limit of its capabilities

could lead to a severe loss in the quality of the service. The following sec-

tions point to the system factors in order of signiﬁcance.

Hard Disks

Speed plays an important role in the caching process, so this factor de-

serves special attention. For hard disks, this parameter is described as ran-

dom seek time, measured in milliseconds. Because the data blocks that Squid

reads from or writes to the hard disk tend to be rather small, the seek time

of the hard disk is more important than its data throughput. For the pur-

poses of a proxy, hard disks with high rotation speeds are probably the bet-

ter choice, because they allow the read-write head to be positioned in the

required spot more quickly. Fast SCSI hard disks nowadays have a seek

time of under four milliseconds. One possibility to speed up the system is

to use a number of disks concurrently or to employ striping RAID arrays.

Size of the Disk Cache

In a small cache, the probability of a HIT (ﬁnding the requested object al-

ready located there) is small, because the cache is easily ﬁlled so the less

requested objects are replaced by newer ones. If, for example, one GB is

available for the cache and the users only surf ten MB per day, it would

take more than one hundred days to ﬁll the cache.

602 25.3. Proxy Server: Squid