Forum Discussion

chrisbirley's avatar
chrisbirley
Copper Contributor
Jan 21, 2025

ASCHI cluster different RAM amounts per node

is it a supported model for ASHCI to have 1 node in a cluster to have different amounts of RAM? i appreciate that you can specify preferred owners, and possible owners on VMs, to restrict/ allow VMs to go to specific hosts, and understand that under normal circumstances you probably wouldnt want to have hosts with different amounts of RAM to circumvent over provisioning and then getting into difficulties upon loosing hosts, but am looking at making one of my ASHCI hosts a SQL server, i dont want to remove it from the cluster due to impacting storage, but need to increase the RAM, and if i can do it to just a single host id prefer that. 

  • luchete's avatar
    luchete
    Iron Contributor

    in my experience, it is possible to have nodes with different amounts of RAM in an ASHCI cluster, but it comes with some caveats. ASHCI generally assumes a balanced resource configuration for high availability and efficient workload distribution, so having nodes with varying RAM could cause issues during failovers or resource contention. If you increase RAM on just the SQL server host, make sure to test how VMs are scheduled and ensure that VM specific resource settings are in place to prevent over committing. Also if possible you have to closely monitor the system to avoid potential problems if the host with more RAM fails. While it’s possible, I’d recommend proceeding cautiously and evaluating the impact on your cluster's performance and resiliency first. Hope it helps! 

    • chrisbirley's avatar
      chrisbirley
      Copper Contributor

      thanks for the response, ive got 2 ASHCI clusters, and my thought process was to use 2 nodes, 1 in each cluster to act as HA SQL, to counter potential host problems, ive got 2TB of RAM in each host, so in general im fine, however ive got a committed 1.9TB of RAM in SQL VMs (and probably an additional 1TB of RAM required going forwards) that i need to migrate to the new clusters (was unaware of this when the ASHCI cluster were bought). i will be stopping all other VMs from being able to use the said host from a compute point of view, and likewise only allowing the SQL VMs to use the specific host, on each cluster, due to licensing. 

      i did think should i just take out the 6th node and make a SQL HA cluster with that, problem is that i loose storage in my ASHCI cluster doing so, and i get too much storage for SQL. so figured upping the RAM on a single host would probably be the better option. 

      • luchete's avatar
        luchete
        Iron Contributor

        Sounds like you're in a bit of a balancing act with the storage and RAM requirements for the SQL VMs. Given that you're locking down the hosts for SQL only, increasing the RAM on a single node could be a reasonable option, especially if it allows you to keep the storage you need for the cluster.

        Just make sure that with the added RAM, the host's resource allocation is carefully managed to avoid any surprises in the event of a failover. Keep in mind that monitoring will be key to ensure everything runs smoothly, and if you have any concerns with failover or VM scheduling, it could be worth doing a test migration to see how the system behaves. Let me know if you how it goes!

Resources