Re: What switches to use?

From: Henrik Nordstrom <[email protected]>
Date: Tue, 12 Jan 1999 22:17:24 +0100

Jonathan Larmour wrote:

> I would hope the former two aren't experimental (ICMP has been in a
> long time, and I thought async I/O was meant to be one of the big
> features of squid 2).

--enable-icmp isn't experimental, but not ideal for all setups. Until
this is controlled from squid.conf it is not a recommended default
option.

The reason why --enable-icmp isn't on by default is that a lot of
network managers considers ICMP ECHO as a evil thing (probably due to
that it is not needed for normal network operation, and has been used in
a number of network attacks). It also adds to the amount of network
traffic generated by your cache.

> And why isn't async I/O enabled by default for all supported platforms
> (resulting from an autoconf test)?

async-I/O is still a experimental feature, and there are only benefits
from it when you have a slow / strict filesystem compared to the network
load.

Some people have reported that it works good on their systems, others
have the opposite experience. Some users have reported that the
filesystems seems to fill up with lost objects when async-io is enabled.

I also have a number of test situations where async-io behaves
strange/odd, but no test situation indicating any failures yet (this is
on a Linux 2.0.X + glibc system).

To summarise on async-io: it needs more testing and probably some
redesigning before it can be considered a non-experimental feature. If
it works for you, good. If it doesn't then please try to identify the
problem in as much detail as possible and report this to squid-bugs.

---
Henrik Nordstrom
Spare time Squid hacker
Received on Tue Jan 12 1999 - 14:31:11 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:44:00 MST