On 2/1/07, amonte <ax.mount@xxxxxxxxx> wrote:
Anyone here use Stripe.All.Mirror.Everything methology? The one claimed by Mr Loaiza 6 years ago?
Yes, it works pretty well for us. Not long ago I posted to list several I/O problems I have with Storage. I
was told yesterday that the EMC we have is using Meta Devices, some sort of SAME. I was told by the EMC guy that the SAN has 120 disks with RAID 1+0, 256GB disks. In the factory it is configured with a internal stripe, 256GB disk is divided into slices of 8GB. So we have 120 disks each disk with 32 slices. With that configuration you take slices from several physical disks to form LUNs. He syas that this is what they call Mete Devices and usually they do that only when the customer is in a hurry and needs the disks badly and dont have time to make a better planning. He further said that this sort of configuration he has seen response times of over 120ms and it is not unusual. The porblem with our disks is that sometimes we get like 80ms responsetime and most of time beteen 25 and 60.
I am not a storage expert, and don't even play one on TV. Even so, this setup seems rather strange to me. By slicing each disk into 32 8Gb pieces, and then creating LUNs from those, it would seem that from a performance perspective you could easily not any better off than when you used JBOD. Let's say you make a LUN of 10 of those 8 gig slices for a medium sized but very busy database. That database is now limited to the IO rate of 10-20 of the 120 disks in your SAN, even if the remaining 100-110 disks aren't being accessed by any other IO requests. If I've messed this up, I'm sure Kevin or someone will straighten me out. -- Jared Still Certifiable Oracle DBA and Part Time Perl Evangelist