NL410 SSD question

Hmm. That is most unfortunate.

> Hmm. That is most unfortunate.

Hi carlilek,

What is the desired use case here?  In other words, if you had the ability to run a "regular" SSD strategy, which one would you choose and why do you desire it on the NL platform?

On your X* nodes (and S* nodes), the use of L3 cache is configurable.  Check the SmartPools section of the WebUI (on recent OneFS releases I believe that's File System -> Storage Pools -> SmartPools Settings).

With respect to NL nodes (and HD nodes), the use of L3 is more nuanced.  There are actually two ways this cache can be used on these node types, and it depends on the density of drives in the system.  For NL/HD nodes with 6 TB drives or larger, a special form of L3 cache, which I'll call "L3 metadata-only", is configured.  This means that the L3 cache is only used to hold metadata blocks, and *will not* hold file data blocks.  This form of L3 cache is mandatory on these node types, i.e. it cannot be turned off, you cannot use other SSD strategies, etc.

On NL nodes with less than 6 TB drives (e.g. 4 TB drives), the L3 cache is "normal" in the sense that it will cache both metadata blocks and certain file data blocks.  You could also decide to use other SSD strategies if you choose, as there shouldn't be the same constraints I mentioned above on these less-dense NL configurations.

For all X* and S* nodes, there is no such concept as "L3 metadata-only".  That is, L3 caches both metadata and file data blocks, and the use of this cache is configurable.

Is this also true for the SSDs in an X410?

When the node is first powered on, they're coming up with L3 Cache enabled by default. There was no option offered on this.

If it is configurable on the X410 nodes, where would we find the options in the interface?

Hi Kip,

I would prefer to ultimately have full L3 caching, since this is (as you may remember) a cluster made up of S nodes and NL nodes, with the S nodes eventually tiering down to the NL nodes. Since we chargeback our users at the same rate regardless of how old the data is, maintaining some level of performance on older data is desirable. That said, after talking to my boss, I think L3 metadata only will serve our needs adequately for this purpose.

I may be able to do a really weird hybrid structure after our next purchase where I have full L3 enabled on 1 tier of S nodes (S200s) for ingest, then tiering down to another tier of S nodes with metadata R/W enabled, then tiering down to the NLs with GNA or L3 enabled. Which raises a rather critical question: Can NL400s with no SSD using GNA be combined in a tier with NLs with L3 metadata-only? If not, this will pose something of a problem...

OK, as long as I can join disparate SSD strategy pools into a Tier, I'm good.

Should probably spawn this out to a new thread, but what we're looking at is going from this:

Performance tier:

10 S200 22HDD/2SSD

9 S200 18HDD/6SSD

7 S210 18HDD/6SSD

NL tier:

13 NL400 3TB HDD

21 NL400 4TB HDD

to this:

Ingest Performance (L3):

10 S200 22HDD/2SSD

9 S200 18HDD/6SSD

Longer term performance (MDRW):

7 S210 18HDD/6SSD

4 S210 23HDD/1SSD

NL tier:

21 NL400 4TB HDD/GNA MDRO

6 NL400 6TB/1SSD L3MD

--Ken

There is one SSD strategy per node pool.  So if have SSD in your NL node pool, it doesn't do GNA, it does L3 (metadata read).  If it has no SSD and you have sufficient SSD on another node pool, you can do GNA.  If you have different drive configs, they will be different node pools.  You can join them for SmartPools purposes as a tier, but that is only for data placement.  With respect to SSD usage, they will follow the rule of the node pool to which they belong.

The expert gave a professional answer, i would like to share you with basic knowledge introduction and configuration demonstration in the thread EMC Isilon OneFS SmartFlash

Adam, can you clarify, how to set "SSD strategy" per node pool?

afais, per node pool [1] we have protection policy and L3 config,

but no choice for the "traditional" SSD stategies (ro, rw, avoid),

as these are set on file pool level [2] with a single global default [3].

[1]

# isi storagepool nodepools view x200_ssd

                 ID: 4

               Name: x200_ssd

              Nodes: 1, 2, 3

  Protection Policy: +2:1

             Manual: No

         L3 Enabled: No

L3 Migration Status: storage

               Tier: -

              Usage [retracted]

[2]

# isi filepool  policies view 1

                              Name: [retracted]

                       Description: [retracted]

                       Apply Order: 1

             File Matching Pattern: Path == [retracted]

          Set Requested Protection: default

               Data Access Pattern: concurrency

                  Enable Coalescer: True

               Data Storage Target: x200_ssd

                 Data SSD Strategy: metadata

           Snapshot Storage Target: x200_ssd

             Snapshot SSD Strategy: avoid

[3]

# isi filepool default-policy view

          Set Requested Protection: default

               Data Access Pattern: concurrency

                  Enable Coalescer: True

               Data Storage Target: x200_ssd

                 Data SSD Strategy: metadata

           Snapshot Storage Target: x200_ssd

             Snapshot SSD Strategy: avoid

Thank you

-- Peter

Yeah, I mis-spoke there (mis-typed, mis-thought?, whatever).

L3 or not L3 is per node pool.  If you are not L3, then you can choose meta read/meta write/data/avoid at the file pool level.

Sorry for the confusion. 

Per-pool defaults for all the protection/SSD-related settings would definitely make (more) sense!