Legacy HDS Forums

The ultimate cache thread

Discussion created by Legacy HDS Forums on Jul 31, 2006
Latest reply on Aug 24, 2006 by Legacy HDS Forums

Originally posted by: Snig

I have been having some discussions internally about cache and the destaging of cache to disk, so I thought I would post some questions here and get the answers from the guru's themselves.  This thread will start with the basics and will eventually walk a write IO though it's lifecycle to the backend disk.  I then want to expand it into striping and discuss software RAID0 over hardware RAID5/RAID1 and go from there.  I told you it would be the ultimate.  ;-)  Pictures or PPT slides would be sweet.

All of this will pertain to a Unix host attached to USP1100 using 146GB/10K drives in a 7+1 RAID 5 configuration.  The Unix host is using 64K stripes for it's file system.  The disk subsystem is carved up into 14GB LDEVs using the OPEN-V emulation.

So here is my first set of questions:

- A cache slot is made up of cache segments.  For an Open-V device how big is a cache slot?  How many segments per slot?

- A cache segment is made up of cache blocks.  How big is a cache segment?  How many cache blocks in a segment?

- A cache block is made of Subblocks.  How big is a cache block?  How many Subblocks are in a block?  How big is a subblock?

- Assuming that a full 64K write comes into cache from the host,is it accepted in as the full 64K or broken into something smaller?  If smaller, how is that broken up?

- How big are the stripes to the backend disk when destaged from cache?

That's good for now.  We'll move on once we have these answers.