Cluster Configuration Best Practices Question

[hooshoo]

I'm not sure if this is the proper board to post this question but I could not find a more suitable one.

Obviously when building out an HA cluster you want the hardware specs (memory, cpu, etc) to be the same, or at least very similar. I have a cluster that consists of HP servers and I want to add more memory in each of the hosts. 3 of the hosts are Gen 7 servers the others are Gen 8. The Gen 7' hardware (memory, cpu) specs are all the same. The Gen 8 specs are all the same and VERY close to the same as the Gen 7 hosts - the only differences is the speed of the memory. 1333 MHz in the Gen 7 and 1666 MHz in the Gen 8s. The Gen 7's have about 20GB less memory than the Gen 8's and the  processors are 2.6 and 2.5 GHz. But, they cluster is balanced and have no issues there.

Here is my question, and this is based on pricing, I can upgrade the memory in my Gen 7 hosts for dirt cheap but they are 1066 MHz vs the 1333 that are in there now (all DIMMS will be replaced). The memory in the Gen 8's will just be in addition to what is already there and they will remain 1666 MHz. Is the big difference in speed (from 1333 to 1066 MHz) a possible issue? Could that cause a unbalanced cluster? Or would it really just be an issue if the CPU and Memory capacity is vastly different?

[Josh26]

This sort of difference in terms of memory speed won't mean anything from the view of your clusters. As long as the CPUs are close enough that you can vmotion between them, it will be fine.

[OscarDavey]

I dont think you will have any issues , its all up to the boards tho , some motherboards need a specific type of ram to be installed  , i suggest that you look that up in the manual first , and in the end and answering your question 1333 is better than 1066 but ... im not sure if its worth the $$ to do such an upgrade now for it .

Best regards

Yours, Oscar

[jrmunday]

If you haven't already done so, you might want to use the HP DDR3 Memory Configuration Tool to confirm the speed based on the number of DIMM's used. I know on the Gen 8's, once you use more than 2 slots per channel, the speed reduces (even with faster ram).

I had a similar scenario recently, and decided to go for more memory (384GB per host) at 1066 MHz rather than 256GB per host at 1033 MHz. The other performance consideration is the number of channels you are using, so in some cases it is better to use smaller DIMM modules and populate more slots to gain the extra channels ... it's a balancing act to find the sweet spot between cost and performance.

As always, faster is better but perhaps the above will provide some more angles to consider in the overall strategy.

Cheers,

Jon

[grasshopper]

Hi hooshoo,

Welcome to the community!  As jrmunday so kindly linked for you, that HP calculator is your best friend when deploying or re-architecting HP servers (save that link!).  You should consider awarding him points by clicking "Correct" since he seems to have answered your query.  In addition, you can also award points for "Helpful" answers as well.  Anyway, I will also provide some feedback based on my experience running VMware on HP below.

As you will see by reviewing the HP calculator, the sweet spot for G7’s is 144GB of RAM which gets you a speed of 1333MHz when using the fastest currently supported memory.  Adding any amount of memory above that will drop you down to 1066 MHz.  For the G8’s, the sweet spot is 256GB which gets you 1600 MHz.  Adding any amount of memory above that will drop you down to 1333MHz.

Personally I always deploy my HP clusters with the fastest possible RAM config (i.e. the “sweet spot”), so my G7s are always 144GB (@1333MHz) and the G8s are always 256 (@1600MHz).  Also, where possible I prefer to make dedicated clusters for the G8s to get the most out of them (i.e. as opposed to dummying them down to lower EVC modes, etc.).  Not always an option though I know.

Anyway, depending on your workload characteristics and tolerance for perceived or actual slowness you will have to decide what’s right for you.  Before intentionally degrading the memory speed (either by using slower memory or missing the sweet spot) you should study this topic well and understand the tradeoffs.  In some cases the requirement for memory capacity outweighs the performance requirements.  That’s a business decision more than anything.  App owners always want the fastest possible so that's what I give them :smileyhappy: