GNA or L3 FlashCache on OneFS 7.1.1

[Henrik Ravn]

Hi

Soon we will have OneFS 7.1.1 and 2% SSD on our Isilon cluster. The cluster is 5 * X410 + 5 * NL400. We can see that the cluster is using some time on metadata  read operations. There is many small files on the cluster (200 mill files).

With OneFS 7.1.1 we can select between pure GNA on the SSD or L3 FlashCache.
If we select GNA the SSD is only used for GNA and some of the SSD can be unused. 

If we select L3 FlashCache all the SSD is used for a mix of Metadata and fil cache.

Anybody having experience with the above?

Thanks and best regards

Henrik Ravn

[Alistair Stewart]

There is some confusion over global namespace acceleration (GNA), what it does and when it is needed.

If a node pool consists of nodes with SSD, they will be used to accelerate the metadata of files in that pool, regardless of the GNA setting. If there are other pools in the cluster without SSDs, the metadata for files in those pools will only be accelerated using SSD from the other pools if GNA is enabled and at least 2% of total storage is SSD and at least 20% of nodes have SSD.

So, if all of your nodes in all pools have SSD then GNA makes no difference.

Finally, if L3 FlashCache is enabled in a pool then *ALL* of the SSD capacity in that pool will be used for that instead of metadata caching.

Al...

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
Regards

Al…

[Peter Serocka]

On 2014 Sep 19. md, at 18:38 st, Alistair Stewart wrote:

There is some confusion over global namespace acceleration (GNA), what it does and when it is needed.

Well, I found this white paper helps out comparing L3 and GNA:

File System Caching Infrastructure

http://www.emc.com/collateral/hardware/white-papers/h13249-isilon-onefs-smartflash-wp.pdf

Henrik,

naturally there is not too much experience

with L3 cache yet, but from my use of

GNA with X200 + NL400 I'd say: 

if your data grows further before the new cluster does,

let's assume those 200 Mio file you mentioned will end up

on the five NL nodes one day. Then without GNA you

might just experience the edge of what's

reasonably doable with NL nodes. In particular

if you rely on large treewalks in regular

workloads or for backups. (i.e. neither using

SyncIQ nor the new snapshot-incremental NDMP mode).

GNA has really saved us over here in such situations!

In case you run into this scenario,

L3 cache on the X410s might still be shining,

but IMHO this would hardly make up for 

NL400s badly lagging behind at the same time.

On the other hand, if you are archiving only

*large* files on NL, while those small files

are mostly active worksets on X410,

the L3 cache may come out fantastic, and

NL nodes are "happy", too.

One good thing is, you can try out both variants

and change them on the fly (whith some time taken 

to re-populate the SSDs, but that's non-disruptive 

to operation.)

-- Peter

Peter Serocka

CAS-MPG Partner Institute for Computational Biology (PICB)

Shanghai Institutes for Biological Sciences (SIBS)

Chinese Academy of Sciences (CAS)

320 Yue Yang Rd, Shanghai 200031, China

pser...@picb.ac.cn

[Jean-Baptiste Denis]

By the way, this document is very well written and explains everything :

http://www.emc.com/collateral/hardware/white-papers/h13249-isilon-onefs-smartflash-wp.pdf

[Henrik Ravn]

Hi

All: Thanks for the reply. 

Peter: Fine input in the discussion. We will go for the GNA and keep an eye open for the level 3 cache. The files(both small and big) will be on both the X-nodes and the NL-nodes, so we will get an improvement when using GNA. It will be interesting to see how much of the SSD that will be used for the GNA (if there is a way to see this).

Best regards

Henrik Ravn

2014-09-19 13:12 GMT+02:00 Jean-Baptiste Denis <jbd...@pasteur.fr>:

By the way, this document is very well written and explains everything :

http://www.emc.com/collateral/hardware/white-papers/h13249-isilon-onefs-smartflash-wp.pdf

[Peter Serocka]

On 2014 Sep 23. md, at 14:08 st, Henrik Ravn wrote:

> Hi
> All: Thanks for the reply.
>
> Peter: Fine input in the discussion. We will go for the GNA and keep an eye open for the level 3 cache. The files(both small and big) will be on both the X-nodes and the NL-nodes, so we will get an improvement when using GNA. It will be interesting to see how much of the SSD that will be used for the GNA (if there is a way to see this).

Wish there was a way... As of 7.0.x, the SmartPools report
doesn't show details for metadata space consumption :(

SSD metadata usage (including the GNA case)
can be quite uneven over time, in particular when
deleting many-small-files under multiple snapshots.

You wrote you have 2% on SSD, that should
leave enough headroom for irregular peaks.
The cluster will send out warnings already at 3/4
of SSD consumption...

Best of luck

-- Peter