Using Intel Optane/DAX/PMEM with FCI

  • I'm working to plan a new upgrade to our existing data warehouse infrastructure.  We're looking at local redundancy options and figure that a traditional FCI will be better than an AG due to the large data sizes we are working with and our San admin's discouragement to doubling our storage costs.

    At the same time, this will be a SQL 2019 Enterprise Environment and we're looking at Optane NVME disks or NVDIMMS to help provide performance improvements while loading data to this environment.

    Besides using the NVME disks for TempDB, is it possible to use local DAX formatted devices for Hybrid Buffer Pool or Persistent Log Buffers along with a traditional Windows Server Failover cluster?

    Has anyone used these?  Can they can be used with a FCI, I can't find anything documented one way or another?  My gut feeling is no, because they are local resources, but maybe?

    Thanks,

    -Luke.

    To help us help you read this[/url]For better help with performance problems please read this[/url]

  • Thanks for posting your issue and hopefully someone will answer soon.

    This is an automated bump to increase visibility of your question.

  • Hey Luke,

    Trying to do something similar,  Would be really interesting to hear if anyone out there has done much with Optane P-MEM and SQL 2019, and what use case they are using - Memory Mode, App Direct, Storage over App Direct.

    For persistent modes can you use FCI - my inclination coincides with yours; there would have to be some sort of replication done since it's viewed as local storage.  Does it require Storage Spaces?  If so that's problematic since SS doesn't play well with SAN.

    I have a new Dell test cluster being installed and will proceed to go through some scenarios, but it would be nice if there was some more data available in the community of anyone actually using/testing Optane with SQL 2019.

    -Tony

  • I checked on #sqlhelp and the general feeling was that it either just wouldn't work (no shared resources) and or wouldn't be supported.

    I was thinking about HBP as a use case with FCI since as far as I know the buffer pool doesn't last past an instance restart and everything needs to be read up into memory again.  Similar to putting tempdb on local ssd with a Failover cluster.  TempDB gets recreated so there is nothing persisted with a failover, as long as the same drive letters are available on

    We had a consultant tell us we could get better throughput for our OLTP system if we switched to some of the Optane based techniques (using PMEM to store the "tail of the log"/Persistent Caching the last 20 MB) and were looking to validate it.  In short, I don't think this works with an FCI and our management folks would rather not have us double our storage footprint to facilitate AG's...  Plus you can't use these with the Primary AG anyhow so...

    I'll be interested to see where this technology goes, but today it doesn't look like an option for my company.

    To help us help you read this[/url]For better help with performance problems please read this[/url]

  • This was removed by the editor as SPAM

Viewing 5 posts - 1 through 5 (of 5 total)

You must be logged in to reply to this topic. Login to reply