On Thu, 2005-06-09 at 11:36 +0200, Raphael H. Becker wrote: > So, just to work around the LBA64 problem, the idea was to recombine > those two "drives" /dev/da1 and /dev/da2 into a RAID (in FreeBSD). > > Two ways to do this: > a) striping > Every access to the logical drive in FreeBSD (/dev/ccd0) will access > both partitions PRT1 and PRT2. Remember: Every single access to ONE > of those two will result in physical disc action for PD2-PD12 and every > access to /dev/ccd0 therefore will result in two accesses for each > physical disc. --> Bad Idea. Perhaps it's because I haven't had my coffee, yet, :-) but why would this be so? If you make your stripe size large enough, you can create a hybrid effect between concatenation and striping---concatenation for smaller files and striping for larger ones. (If a file is smaller than the stripe size, access will go only to the drive on which that stripe resides. However, if it is larger, then access will be spread over multiple drives, improving load.) I am using geom_stripe on two 80 GB PATA drives to create a 120 GB stripe across both. (I only use part of each drive, and use the other part for a geom_mirror.) I have my stripe size set at 1 MB. I've been very happy with the performance. I used raidtest (now in ports) with different stripe sizes beforehand to try and ascertain a good stripe size. In my case, performance rose steadily and noticeably up to a 1 MB stripe size, and then was only slightly better as I increased thereafter. (I stopped at a 16 MB stripe.) Of course, this depends upon the kind of workload your system is expected to have. Additionally, in your case, you have the interaction with the hidden RAID 5 and whatever stripe size it is using. That possible interaction alone might be enough to use only concatenation, rendering the situation the same as using a large RAID 5. Cheers, Paul. -- e-mail: paul_at_gromit.dlib.vt.edu "Without music to decorate it, time is just a bunch of boring production deadlines or dates by which bills must be paid." --- Frank Vincent ZappaReceived on Thu Jun 09 2005 - 13:36:31 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:36 UTC