Re: Bake-off

From: Glenn Chisholm <glenn@dont-contact.us>
Date: Thu, 8 Apr 1999 01:06:07 -0600 (MDT)

> The bake-off did not have any test cases with high end
> hardware for squid. I would like to see the results of squid
> with decent hardware. Because of the Unix FS, you really need
> 10000 RPM drives, and would be better off and cost less with 3
> 10,000RPM drives (6 10000RPM drives even better) then the 6
> cheap drives it was tested with in the bake-off. Also, a
> $3500 switch and a single gigabit port along with squid
> running on the gigabit port. Just to see how squid "scales"
> when you put decent hardware behind it. It will bring the
> cost closer to $10,000, but wil give a better idea of how
> squid "scales" compared to the others. A low end configuration
> of course should still be tested.... but I think the low end
> would be better off with 2 7600RPM IDE drives.
>
>
All this is likley to do is destroy the price performance ratio for squid.
It might get you a few more requests per second, but not the kind of
change to keep the price performance ratio where it is.

The problem was just not the hardware, it is the file system. a custom
file system is really what is needed to speed us up.

Remember that at 75 requests/sec we are using about 5 mega bits/sec. More
than enough to satisfy the needs of most users.

As Alex has said work has started on a SquidFS which should clear that
bottle-neck and then all of a sudden we will be attempting to correct
another "UNIX Problem" :)

glenn
Received on Thu Apr 08 1999 - 00:52:41 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:45:45 MST