On 07/29/2004 02:20 PM, Si Jones wrote:
A lot of people are saying mirrors are faster for read, and stripes are faster for write, which is mostly true. BUT, if you have, say, a database with zillions of tiny chunks of data, mirrors may be faster for write as well.Hi All,
Could someone tell me which one is faster out of mirror or striped software raid in FC, in there experience?
Cheers
Simon Jones
The issue is that seek time is generally faster with mirrors, because you just read the first copy you find; and stripes write smaller portions of the file simultaneously, so a stripe set of five disks, for example, can write a big file in roughly 1/5 the time (if your bus is fastest enough). But when you're writing a very small chunk of data, the seek time is proportionately more significant so for a tiny write job, a mirror set can outperform a stripe set.
So, at least theoretically, mirrors can be faster for writing in some cases. Whether that is often, rarely, or never true in the real world, I can't tell you. I have had Oracle gurus tell me they use mirrors specifically for this reason for the right kind of data. I don't know if they're right. All I know for sure is, we use RAID 5 for everything because it's safer, and pretty much any disk is faster than the network :-).
--Matt