Re: Poor hit/transfer ratio

From: Alex Rousskov <>
Date: Wed, 3 Dec 1997 10:03:26 -0600 (CST)

On Wed, 3 Dec 1997, Ken Piotto wrote:

> Does anyone know if there is a magic number of users (or better, requests
> per unit time) in order to see a respectable hits to kbytes transfered
> ratio?

I doubt there is such a number. I suspect that both hit ratios depend more
on your traffic patterns rather than on the number of users. Consider for
example leaf proxies versus top level caches. In spite of higher traffic
volumes and intensity, top level caches often have lower hit ratios.

Of course, if you hit extremes like having only 20 users, you should be
very careful with the figures you are getting. A single "weired" user
requesting (for whatever reason) the same page with many small gifs 5
times could skew your statistics significantly.

If you have time, consider writing a script that will filter out N out of
20 users from your log files and _then_ calculate "imaginary" hit ratios
on your own (you cannot trust HIT/MISS fields in the log anymore). This
way you can see how hit ratios changes with N and if it becomes stable

At any rate, I doubt there is any guarantee that your 20 users are, on
average, the same as the _rest_ of your user population. 10000 students vs.
20 professors or 20 hackers vs. 10000 "surfers": the volume of porno
traffic per person will differ substantially, I guess. :)

> Seems the only hits occur on relatively small objects.

This is very common. Usually, more than 50% of all objects are less than
3KB. These determine your document hit ratio. Unfortunately, to get a
decent byte hit ratio, you have to cache a lot of large objects which are,
on average, not popular.

My $0.02,

Received on Wed Dec 03 1997 - 08:09:14 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:49 MST