Quote:
|
Originally Posted by engstrom I disagree with that statement. RAID5 writes are expensive. Cache helps mitigate the overhead but you're still going to pay at least a little penalty. You have to re-read part of the blocks that aren't being rewritten to recalculate the new parity. Large sequential writes on a defragmented disk are an optimal situation (and often what you do when you're editing) but still I never manage to keep my disks unfragmented for long.  |
I didn't say anything about writes. Editing should be read only, with writes going to a local disk/ cache. I specifically said 'throw out data' i.e., read
Because you can interleave the accesses, you can read data faster than from a single disk. Also, conceptually, you have 2, 3 or 4 times the amount of cache available for writes, so there should be some improvement there as well (assuming one disk or a raid array of the same disks)