Re: [squid-users] Solaris's Network Cache Accelerator and Squid

From: Adrian Chadd <adrian@dont-contact.us>
Date: Fri, 5 Apr 2002 21:37:13 -0700

On Fri, Apr 05, 2002, michael.speer@sun.com wrote:
> Joe:
>
> Network Cache accelerator enables in kernel http serving with callout to user
> level processes when a file in not in the in kernel cache. It significantly
> boosts web serving performance, it possibly could do the same for squid,
> since squid needs to delivery web content via http. An API and document for
> this feature of Solaris 8 is available on the Sun website.
>
> I just trying to find out if the squid community was aware of its existence and
> whether anybody was interesting in making squid run ontop of it.
>

Its in the "would be nice, but there's no way to tie squid into it
now" bucket. Sorry.

_although_, this is way too specific an API to be useful for anything
but specific models of http servers. How about you go and hit up
Sun for a public API allowing user-land programs to perform async
socket IO in a posix AIO fashion, where the -kernel- takes the
userland buffers and copies in/out of them directly?

Solaris has support for posix AIO, but a) its a hybrid user/kernel
implementation using threads, and it sucks for lots of busy network
IO (i've tried it), and (b) you still need a poll-type API to get
the completions. FreeBSDs AIO implementation is probably the glosest
to whats needed (kernel-side AIO routines have bypasses for sockets
to speed up network IO, and have aio_waitcomplete() which returns
the next completed aiocb, so you don't need to poll lots and lots)

That'll be a good start that even squid will be able to take advantage
of shortly. :)

Adrian
Received on Fri Apr 05 2002 - 21:37:14 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:07:29 MST