Comentários (15)

1 NigelMorton comentou às Link permanente

This had been driving me spare! Thanks for blogging this. <div>&nbsp;</div> Oh, and I don't have any hot data either.

2 anthonyv comentou às Link permanente

Thanks for the comment, glad it helped. <br />

3 GregS comentou às Link permanente

Hi Anthony, How do I generate a heatmap for an SVC cluster running v6.2.0.3?

4 anthonyv comentou às Link permanente

So to get a heatmap you need easytier to be active (even if you have no SSD). <br /> It would turn on automatically if you added SSD... but I am presuming you dont have it. <br /> First up you need to SSH to you SVC or Storwize V7000 and turn easytier on for every pool. <br /> Do an: svcinfo lsmdiskgrp <br /> To get the pool numbers of each pool. say you have pools 1 and 2. Now do <br /> svctask chmdiskgrp -easytier on 1 <br /> svctask chmdiskgrp -easytier on 2 <div>&nbsp;</div> Now wait 24 hours (could be less). <br /> Now go to Troubleshooting (the monitor icon) --&gt; Support <br /> Click 'Show listing' <br /> Look for a heatmap called something like: <div>&nbsp;</div> <div>&nbsp;</div>

5 GregS comentou às Link permanente

Hi Anthony, That worked fine. I got a heatmap for one of my SVC Clusters. What defines whether a portion of a VDisk is "hot" or "cold"?

6 anthonyv comentou às Link permanente

Hi Greg. <div>&nbsp;</div> Hot data is data that receives high levels of random IO (versus sequential) and small block IO. <br /> Small random IO (reads and writes) are the least spinning disk friendly patterns, so the block that are getting hit with this are better placed on SSD.

7 culdenickel comentou às Link permanente

Hi Anthony, <div>&nbsp;</div> I just found out this blog, thanks a lot for all the useful information. <div>&nbsp;</div> I have a question about the Easy Tier during backup for a volume that is fragmented? Will the high activity will move that volume on SSD tier every time it's being backed up, or there's a kind of protection for that. Or maybe even a fragmented volume is considered sequential IO when backed up? <br />

8 anthonyv comentou às Link permanente

Its a good question. <br /> In general, running one backup is not going to push all the backed up data into SSD regardless of how fragmented the data is at a file system level. Backups tend to still be big block, reasonably sequential and a not repeated pattern of I/O, so not a candidate for hot data.

9 MartinHitchman comentou às Link permanente

I have pulled the DS8700 STAT code from here,, but have found it stalls on execution. has anyone else come across this. <br /> many thanks Martin

10 Me&MySVC comentou às Link permanente

Hi Anthony, <div>&nbsp;</div> I see a 0x0000 Volume ID, which I can not find in the GUI or CLI. This holds the most hot data (3,5G from total 147G)for my environment. Do you know what this is, or how I can investigate? <div>&nbsp;</div> Thanks in advance.

11 anthonyv comentou às Link permanente

So the volumes are in hex. Of course 0x0000 is just volume 0 so that bit is easy. <br /> To work out which volume that is, you need to open a command line window and do the following command: <div>&nbsp;</div> svcinfo lsvdisk <div>&nbsp;</div> Look for volume ID 0.

12 Me&MySVC comentou às Link permanente

Hi Anthony, <div>&nbsp;</div> thanks for the above hint. I think I need to review that list again with the new info. <div>&nbsp;</div> By the way, is there a certain value, where you would decide to add SSD`s? In my way 3,5G from total 147G hot data seems only a bit, but I have those values in several volumes. <div>&nbsp;</div> Is there any official document from IBM or at least an advisor to interpret the results? <div>&nbsp;</div> Thanks in advance!

13 anthonyv comentou às Link permanente

Well there are lots of variables here and hot data is just one. <br /> While 147 GB hot data is enough to consider a RAID1 pair of SSDs, the question is... is disk I/O performance a bottleneck? You may speed up disk I/O very nicely but find that your hosts are still memory or LAN constrained. <br /> I hate to give the classic IT answer: It depends.... but well... it kinda does. <br />

14 aixfplwiz comentou às Link permanente

Anthony - great blog . Great website. so my question would be around disk "re-allocation" or God forbid - "disk (volume) Defrag" - is there a command that can be run so that the V7000 looks at how data is laid out (on a volume by volume basis) and would calculate a "re-org" - Thanks In advance.

15 avandewerdt comentou às Link permanente

Great question - short answer - no there is not. <br /> If you using fully allocated volumes, then de-frag needs to be done at the LVM level. <br /> If you using Storwize V7000 Thin Provisioning, then you could indeed get very fragmented volumes at the disk level, but there is no way to say how non-sequential the data has ended up.