On Wed, Aug 09, 2006, James_Gray@xxxxxxx wrote: > Out of curiosity, has anyone done benchmarking of squid using a formatted > disk partition for cache, versus raw IO (using Veritas Volume Manager > etc)? I've seen massive improvements in RDBMS applications > (Oracle/DB2/etc) when using raw IO, so I'm curious if this is even > possible with squid. I'm offline ATM so googling isn't really an option, > but if anyone has links to any case studies etc with Squid, I'm very > interested. COSS, which Steven Wilton and I have put a lot of effort into over the last few months, is a 'raw io' filesystem which speaks direct to a block device (or file if you prefer.) When benchmarking COSS I noticed using ext3 rather than the raw device gave a rough 20% increase in disk IO bandwidth. I was also seeing the IO being scheduled in 'clumps' rather than evenly over time. So yes, rawio is faster. :0 Adrian