Async I/O Questions

We are currently running without Async I/O on two Sun Ultra 150's with
Solaris 2.6. Squid is configured with --enable-dlmalloc,
--enable-cache-digests --disable-ident-lookups. It's compiled with GCC
2.8.1 with -O3 -mcpu ultrasparc in addition to the standard CFLAGS.
cache_mem is 128MB and cache is 4.9GB accross three 2.1 GB drives with
fastfs. Usage averages 200 to 300 requests per minute on each system.

1. With this configuration is there any performance advantage with async
i/o?

2. Without async i/o when squid gets busy the client either waits until
squid returns the object or times out. What happens with async i/o if
there are not enough threads available?

3. When building with --enable-async-io are the number of threads set
solely by NUMTHREADS in aiops.c or do I also need to compile with
-DNUMTHREADS= in CFLAGS?

4. Are there other compiler optimizations besides -O3 and
-mcpu=ultrasparc for GCC 2.8.1 for running on Sun ultrasparc cpu's and
Solaris 2.6?

TIA

-- 
Jim Richey
jrichey@highmark.com
Highmark, Inc.
http://www.highmark.com