Comments on: Living Loving MAID http://storageadvisors.adaptec.com/2006/04/03/living-loving-maid/ Storage Solutions for Real World IT Professionals Fri, 29 Aug 2008 01:55:55 +0000 http://wordpress.org/?v=1.5.2 Comment on Living Loving MAID by: Tom http://storageadvisors.adaptec.com/2006/04/03/living-loving-maid/#comment-245 Tue, 04 Apr 2006 17:22:56 +0000 http://storageadvisors.adaptec.com/2006/04/03/living-loving-maid/#comment-245 Copan does some tricks with the initial write where they're able to "assume" that the data on the unused drives is zero, and therefore they don't need to do a r/m/w. But they are certainly less clear what they do after the first disk fills up. I agree that the total throughput will be less than that of a single disk. If they do their algorithms PERFECTLY, then they might get half a disk's throughput. Maybe that's about 40MB/s? 40MB/s is definitely less than the throughput of a high-end tape, but maybe Copan's customer's don't care. Anybody out there actually use one of these things? TT Copan does some tricks with the initial write where they’re able to “assume” that the data on the unused drives is zero, and therefore they don’t need to do a r/m/w. But they are certainly less clear what they do after the first disk fills up.

I agree that the total throughput will be less than that of a single disk. If they do their algorithms PERFECTLY, then they might get half a disk’s throughput. Maybe that’s about 40MB/s?

40MB/s is definitely less than the throughput of a high-end tape, but maybe Copan’s customer’s don’t care.

Anybody out there actually use one of these things?

TT

]]> Comment on Living Loving MAID by: ming zhang http://storageadvisors.adaptec.com/2006/04/03/living-loving-maid/#comment-243 Tue, 04 Apr 2006 16:58:00 +0000 http://storageadvisors.adaptec.com/2006/04/03/living-loving-maid/#comment-243 if this is the maid algorithm, then this speed is not decent at all. most time virtual tape backup come with long sequentical stream write. so if u write A B C D. in raid4/5, it is a full write and u need not read parity at all. so it can archieve 4x single disk speed here. but for maid, it will do r-m-w for each block. it will be VERY slow. start from LBA X, assume the block size is y, so u need to read Y from disk at X, seek back to X, write Y to disk at X. (luckily here that next read will start from X+Y so save a seek here) . i do not have a number here, but i would say this is much slower than a sequential throughput u can get from a single disk. so how could this to be decent? if this is the maid algorithm, then this speed is not decent at all.

most time virtual tape backup come with long sequentical stream write. so if u write A B C D. in raid4/5, it is a full write and u need not read parity at all. so it can archieve 4x single disk speed here.

but for maid, it will do r-m-w for each block. it will be VERY slow. start from LBA X, assume the block size is y, so u need to read Y from disk at X, seek back to X, write Y to disk at X. (luckily here that next read will start from X+Y so save a seek here) . i do not have a number here, but i would say this is much slower than a sequential throughput u can get from a single disk.

so how could this to be decent?

]]>