Storage Boards Cleanup
To make it easier to find information about HPE Storage products and solutions, we are doing spring cleaning. This includes consolidation of some older boards, and a simpler structure that more accurately reflects how people use HPE Storage.
3PAR StoreServ Storage
cancel
Showing results for 
Search instead for 
Did you mean: 

Reclaim space, problem with chunklet links?

owiking
Occasional Advisor

Reclaim space, problem with chunklet links?

Hi

As many others we are having trouble reclaiming space in our 3Par 7200c.
We are running ESXi 6 and some physical Windows servers.

We had higher hopes on the deduplication ratio on the 3Par, (we are at 1.4-1 with 16kb cluster size) but we have gotten used to it being low so it doesnt bother us so much anymore.

But because of the low ratio we bought 2 cabinetts with FC drives. We wanted to move servers to those drives that doesnt need the speed of the SSD drives. About a month ago we moved about 6.5TiB data to the FC drives so that we could get more free space on the SSD drives.
We moved virtual servers in our VMware environment. We have been running unmap and everything else one can think of (compact CPG, Tune System), but after more than a month we only got back about 1TiB when we wanted to get back 6.5TiB.
The 3Par console shows the disks that we freed to be 0% used, but still when highlightning the SSD CPG we only see that we have gained back 1TiB.
We decided to delete one of the virtual volumes and force 3Par to give us back the space, but even after deleting a volume we knew had contained 1.5TiB data we got nothing back.

We are starting to think there might be some problems with the chunkstore. I heard somewhere that there could be a bug in the chunkstore, something about the chunkstore not deleting unused chunks. Anyone know about that?

About reclaiming space, has anyone found any difference in the amount of space you get back and the time in which you get it between running Thinly deduped and Thin Provisioning?

We are thinking about moving away from using Thinly deduped and only using Thin Provisioning in hopes that we shall be able to reclaim space easier and also know how much space we have left.
But we cant go that way as it is now because moving data to Thinly provisioned disk and not getting anything back from the deduped volumes makes us run out of space.

Thanks.

 

5 REPLIES
yapkl
Valued Contributor

Re: Reclaim space, problem with chunklet links?

Hi owiking,

Do you have these info?

1. 3PAR model
2. 3PAR OS version
3. SSMC/IMC version
4. Serial Number of the 3PAR (please PM if you do not wish to reveal publicly)
5. How many CPGs do you have configured?

On ESXi you need to run commands to reclaim the free space. Check out this whitepaper, in page 42 it talks about the esxcli unmap command to run.

https://www.hpe.com/h20195/v2/GetPDF.aspx/4AA3-8987ENW.pdf

The other thing with TDVV is after you've done some space reclamation, if you want to see statistics immediately you need to run updatesnapspace -dedup_stats, refer to page 26.

 

 

 

Best Regards,

yapkl

Note: The opinions & comments above are my personal opinions, not of Hewlett Packard Enterprise
----------
If my post was useful, click on my KUDOS! thumb below! Thanks.
Gary_L
HPE Pro

Re: Reclaim space, problem with chunklet links?

Hello,

We are freeing space to be reused, 16K pages as they are unused and the collection runs we free those back up to be reused within the DDS.  We also reclaim back to the CPG at 128MB of contiguous space.  Based on the way data is written to the DDS the space being reclaimed from the DDS may be minimal.

If you have the space to convert all VV's to TPVV you could reclaim all of the DDS space after the last TDVV is removed.  This is all about space management considering you would need the space to convert all VV's before realizing the DDS capacity being reclaimed.

After this you could use the dedup estimator to identify dedup estimates on combined volumes to see what would be beneficial to move back to TDVV.  Based on 1.4:1 dedup ratio on the CPG, (that's the only place you should be looking at the dedup ratio, showcpg -s), it's not the worst but not the best it could be and that is all dependent on the types of data on disk.

Gary

I am an HPE Employee
owiking
Occasional Advisor

Re: Reclaim space, problem with chunklet links?

Hi

I must say that Im surprised by your answer, if Im reading it corectly then you can never get back space on TDVV's, except in small numbers without deleting all of the TDVV's and thus clearing the dedup data store?

Do you know if HPE plans to implement defrag on the DDS on 7200c systems?

We bought FC disks to be able to free up SSD space, but since we don't get any space back on the SSD that uses TDVV's then we are back were we started, only that we are now using disks with less performance. 

I feel cheated, who wants to buy a storagesystem where you can't get any space reclaimed without deleteing all TDVV's? We dont have 25-30TiB of free space laying around.

If we had known this before we would have never used the inline deduplication at all. 

Thanks for your answers.

 

3Par model: 7200c

3par OS version: 3.2.2 MU2

We have 2 CPG's. One for SSD RAID5 and one for FC RAID5.

Gary_L
HPE Pro

Re: Reclaim space, problem with chunklet links?

Hello,

We are freeing space to be reused, 16K pages as they are unused and the collection runs we free those back up to be reused within the DDS

At 1.4:1 you are still achieving space savings of 40% so the returns are there for continued usage.

Gary

I am an HPE Employee
owiking
Occasional Advisor

Re: Reclaim space, problem with chunklet links?

Hi

I saw you changed the earlier post, where you said "With 7K arrays there is no defrag on the Dedup Data Store, (DDS)". I was worried about that but good that defrag of the DDS should'nt be a problem then.

I think I'll create a case with HP and see if they can help me with the problem of getting free space back.

Thanks.