Is there a CPU impact when adding a flashpool to a system (metrocluster)?

[gmalego]

Hi all

I have a question related to flashpools and the CPU impact this might have:

When adding a flashpool to a system, will this impact the overall CPU usage compared to adding a regular non-flashpool aggregate ? Is there difference in CPU overhead when adding a flash cache vs a flashpool ? We are looking to add a flash pool to a 3250 MetroCluster config, of which the CPU is already at 50% per node, and want to avoid that in takeover the system becomes barely usable after adding the flashpool and moving workloads towards this new mirrored hybrid aggregate. 

Many thanks in advance for the feedback,

Best regards,

Gerrit

[sshapiro · Accepted solution]

Hi Gerrit,

Flash Pool caching does use more CPU than Flash Cache.  Inserting data into Flash Cache is essentially a buffer cache copy operation. Inserting cache into Flash Pool cache is a write operation to the SSDs, the same process that is used to write data to HDDs.  In addition, Flash Pool cache metadata (e.g. the cache heat map) is storage on the SSDs; that enables Flash Pool cache to remain available and consistent during controller takeover events. If the Flash Pool aggregates are properly Pool Cache versus using Flash Cache is typically not significant in terms of performance impact or overhead.  If you are a partner, you can use SPM (System Performance Modeler) to see the difference in controller utilization between Flash Cache and Flash Pool.

There is a very good reason to use Flash Pool instead of Flash CAche in a MetroCluster system. The Flash Pool caches in each plex are kept in sync; if a failover occurs, the cache on the remote node is already warm with data from its partner's volumes. That is not the case with Flash Cache in a MetroCluster system - the cache has to be warmed with data from volumes from the node that went offline.

Skip