Re: Raw vs. cooked

Logging could also make a difference - i.e buffered v/s unbuffered. Under unbuffered logging, the 'raw' benchmark would 'suffer' because logging would be truly unbuffered, where as the 'cooked' benchmark would benefit from the FS buffering inherent in Unix. Which also highlights the advantage of Raw chunks when using unbuffered logging - log records are guaranteed to be written. Rudy In article <3739862d.137464473@news.maricopa.gov>, fprose@rocketmail.com (FProse) wrote: > You can do your math any way you want to -- I do mine with a stop > watch!. The bottom line was the cooked ran faster on a 1+ hour > benchmark. > > BTW, it was raw disks, 5 db spaces, one per spindle. > > --== Sent via Deja.com http://www.deja.com/ ==-- ---Share what you know. Learn what you don't.---

Rudy Fernandes wrote: > Logging could also make a difference - i.e buffered v/s unbuffered. > Under unbuffered logging, the 'raw' benchmark would 'suffer' because > logging would be truly unbuffered, where as the 'cooked' benchmark would > benefit from the FS buffering inherent in Unix. There is no file system buffering even when cooked file system chunks are used. As earlier posters mentioned the O_SYNC flag (see 'man fcntl') or other operating system functions are used to make sure that data is immediately flushed to the disk. Regards, Heiko > > > Which also highlights the advantage of Raw chunks when using unbuffered > logging - log records are guaranteed to be written. > > Rudy > > In article <3739862d.137464473@news.maricopa.gov>, > fprose@rocketmail.com (FProse) wrote: > > You can do your math any way you want to -- I do mine with a stop > > watch!. The bottom line was the cooked ran faster on a 1+ hour > > benchmark. > > > > BTW, it was raw disks, 5 db spaces, one per spindle. > > > > > > --== Sent via Deja.com http://www.deja.com/ ==-- > ---Share what you know. Learn what you don't.---