Stripe dirty bitmap
Sure would be nice to have md track whole never written stripes, so creating an array didn't have to be done over a three day weekend. When any part of a virgin stripe was written it could then be initialized properly, and when a virgin stripe was read zeros could be returned without disk i/o. It doesn't matter that the reads are done in parallel, with modern disk the bus is the bottleneck. With eSATA arrays the bus is slow to start with, passing TB through it should be avoided if possible.
Signature · 3 lines
Bill Davidsen <[email protected]> "We can't solve today's problems by using the same thinking we used in creating them." - Einstein
Quoted 8 lines · On Sun, 04 Jul 2010 20:26:40 -0400 Bill Davidsen <[email protected]> wrote:
Sure would be nice to have md track whole never written stripes, so creating an array didn't have to be done over a three day weekend. When any part of a virgin stripe was written it could then be initialized properly, and when a virgin stripe was read zeros could be returned without disk i/o. It doesn't matter that the reads are done in parallel, with modern disk the bus is the bottleneck. With eSATA arrays the bus is slow to start with, passing TB through it should be avoided if possible.
Sure would.
NeilBrown
Quoted 10 lines · Neil Brown
Quoted 3 lines · On Sun, 04 Jul 2010 20:26:40 -0400 Bill Davidsen <[email protected]> wrote:
Sure would be nice to have md track whole never written stripes, so creating an array didn't have to be done over a three day weekend. When any part of a virgin stripe was written it could then be initialized properly, and when a virgin stripe was read zeros could be returned without disk i/o. It doesn't matter that the reads are done in parallel, with modern disk the bus is the bottleneck. With eSATA arrays the bus is slow to start with, passing TB through it should be avoided if possible.
Sure would.
Have to have a copy on each drive, I think? Use a bit for a group of N stripes or one per stripe? Put N in the metadata? I'd use stripe_cache_size, but that changes.
As disk gets cheap (1.5TB for $65 this weekend?) bus speed is being an issue, more so with software raid which does the io out of system memory.
Signature · 3 lines
Bill Davidsen <[email protected]> "We can't solve today's problems by using the same thinking we used in creating them." - Einstein