Quoting Leonid Podolny, from the post of Wed, 06 Apr: > > After reading my own letter, I think that my point is not very clear. > When I run the same test on 16 separate disks (i.e. no RAID) all the > disks work at about 90% and the overall result is significantly (~70%) > better. Thus, it looks like software raid ruins the performance instead > of improving it.
well, if you are writing more than one copy of each sector, of course you'll see a drop in performance, what did you think? RAID1 is slower than no RAID, but RAID0 is faster. the reason is that in RAID1 you double the amount of writes, whereas in RAID0 you balance the number of reads and writes between more than one disk, however that loses some reliability. the solution many people choose in such a case is RAID 1+0, at times known as RAID10 or RAID0/1 and other names. it means either creating two mirror pairs, striping them and using this as the logical disk, or taking two striped blocks of two disks each and have them mirror eachother. either way, it adds both redundancy and balances hardware access, so the bottom line looks closed to single-disk speed. -- Evil Kneivel's role model Ira Abramov http://ira.abramov.org/email/ ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
