Element Size Question:

Question

Does anyone know of a good source that details a write operation within the Clariion?

Specifically, If I have a RAID5 (4+1) with the default element size of 64K, the stripe size is 256K...

What happens if I write a file that is smaller than 64K (say 8K?)

Is 8K written to the first drive or is it broken into 4 2K segments and written to all 4 drives?

Also, when exactly is parity written? I would assume it is after each and every write operation, but how does that work in the above example of a 8K write?

Thanks!

DGM3 · Accepted Answer

Q2: This is what I suspected... So, does the Clariion wait until the next 64KB element is full before it starts writing to the next drive or is the next 8KB I\O written to the next drive in the RAID set since it's a new I\O request?

A2: It's not the case that elements are "filled". In the context of block storage devices that present a linear address space (that is, a LUN that spans from block 0 to block n), the placement of data is determined by the host (block storage devices have no concept of files). See for example the RAID 5 diagram here;
http://en.wikipedia.org/wiki/Raid_5#RAID_5
If each element is 128 blocks in size, host visible blocks 0-127 reside in A1, host visible blocks 128-255 reside in A2, etc. If the host issues an 8KB write directed to block 200, those applicable blocks within A2 will be directly written, independent of where other IOs may be directed.

Q2: So I take it that if 8KB is written to only 1 drive and it partially fills the element, the parity data is also cacluated and written to a different drive (partially filling another element?)

A2: For any write operation, corresponding parity locations within the stripe are updated. Thus in the example above; an 8KB write directed to block 200 (which resides in A2), will also result in 8KB worth of parity within Ap being updated. Had the write gone to locations within Dsomething, then parity within Dp would have been updated, etc.

Regards,
DGM

kelleg · Answer

On PowerLink:http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/H2358_clariion_best_prac_fibre_chnl_wp_ldv.pdf?mtcs=ZXZlbnRUeXBlPUttQ2xpY2tTZWFyY2hSZXN1bHRzRXZlbnQsZG9jdW1lbnRJZD0wOTAxNDA2NjgwMWI1NzZlLGRhdGFTb3VyY2U9RENUTV9lbl9VU18wThis is the Best Practices guide and it has a section about element size.glenk

Bowling1 · Answer

Thanks for the link.

I have read that one in the past and agree that it's the best out there, but unfortunately it leaves me with more questions.

I'm really curious how data gets distributed across many disks in an environment that has small file I\O sizes.

Say a RAID 5 (8+1) vs. RAID 5 (4+1). Since the (8+1) stripe set is 512KB and the (4+1) is 256KB, does the smaller RAID set make more sense in this environment (or does it really matter at all?)

(If my element size is 64K but I only have 4KB to write, when\how is the data distributed across the RAID set and when is parity calculated?)

Thanks

DGM3 · Answer

Q: What happens if I write a file that is smaller than 64K (say 8K?) Is 8K written to the first drive or is it broken into 4 2K segments and written to all 4 drives?

A: In the context of RAID 5, the 8KB will be written just to the first drive. The RAID level that will write data "across" all drives in parallel is RAID 3. The above are standard RAID behaviors that are not specific to CLARiiON.

Q: Also, when exactly is parity written? I would assume it is after each and every write operation, but how does that work in the above example of a 8K write?

A: Parity is updated (written) essentially at the same time the data is written, however there are multiple disk IO operations involved. Standard RAID 5 writes (again, not specific to CLARiiON) involve 4 steps; read old data, write new data, read old parity, write new parity. The latency of performing these operations is often not visible to hosts, due to Write Caching on the array.

Regards,
DGM

DGM3 · Answer

Q: Say a RAID 5 (8+1) vs. RAID 5 (4+1). Since the (8+1) stripe set is 512KB and the (4+1) is 256KB, does the smaller RAID set make more sense in this environment (or does it really matter at all?)

A: In the context of a single write operation, it's not going to matter. The question to consider is how many total IO/sec will host(s) be directing at the RAID Group? The more disks that are configured in the RAID Group... the more total IO/sec the RAID Group can sustain. The Best Practices document discusses how to determine how many drives should be configured in a RAID Group, based on various factors such as the expected host IO loading and read/write mixture.

Please note that changing the element size from the default setting is strongly discouraged (adjustments to it typically cause more harm than good).

Regards,
DGM

Bowling1 · Answer

DMGVery good. Now we are getting somewhere...A: In the context of RAID 5, the 8KB will be written just to the first drive. The RAID level that will write data 'across' all drives in parallel is RAID 3. The above are standard RAID behaviors that are not specific to CLARiiON.Q2: This is what I suspected... So, does the Clariion wait until the next 64KB element is full before it starts writing to the next drive or is the next 8KB I\O written to the next drive in the RAID set since it's a new I\O request?_______________________________________________________________A: Parity is updated (written) essentially at the same time the data is written, however there are multiple disk IO operations involved. Standard RAID 5 writes (again, not specific to CLARiiON) involve 4 steps; read old data, write new data, read old parity, write new parity. The latency of performing these operations is often not visible to hosts, due to Write Caching on the array.Q2: So I take it that if 8KB is written to only 1 drive and it partially fills the element, the parity data is also cacluated and written to a different drive (partially filling another element?)Thanks Again!

Bowling1 · Answer

Thanks again DMG...Unfortunately, I can't get to wikipedia - Blocked - Got to love technologyI'm sending off a request to have the site unBlocked, but the request takes 2 weeks (Silently Sitting Shaking my head)...I'll review it at home tonight, but it sounds like what I'm looking for. My understanding was incorrect. I was under the impression that the stripe set was composed of 128 Blocks (64K elements) per disk and the Clariion tried to fill each element before going to the next disk - I couldn't believe that would be an efficient model in a small IO write environment.Thanks again. This really helps.

kelleg · Answer

Try a seach on PowerLink for the following:

White Paper: EMC CLARiiON Fibre Channel Storage Fundamentals - Technology Concepts and Business Considerations

Has a pretty good explanation of element sizes and how they are used.

glenk

CLARiiON

Element Size Question:

Was this post helpful?