Understanding CPU aggregation from one generation to the next
The viewer reports aggregation of CPU alongside the IOPS, MB/s, and Memory on the Aggregation Summary page itself.
This will allow you get a better indication of migrating older server CPU generations to newer generations.
By abstracting the type of CPU itself and just looking at the aggregate cycles used you can do the math in a hardware or generation agnostic approach.
For Example:
----------------------------------------------------------------------------
If your report show all of the servers combined are utilizing 348GHZ of processing power. You will need to accommodate for that in your new design.
If i was looking to move to server platform with 2.0GHZ 8 core processors, I would need to have at least 11 servers to be able to support the workload (11servers x 2processors x 2ghz x 8 cores) = 352ghz.
That would be just above my requirement threshold. One would make adjustments at that point to accommodate any growth forecasts.
-
On this same subject I have had a mail from the a customer looking for guidance on the difference between CPU generations. We understand that DPACK aggregates the total GHz of processing power, but a cycle of an E5 v2 CPU for example would not be comparable to a cycle of Skylake SP CPU.
--------------------------------------------------
Customer Mail :
I am keen to understand how we can normalise the peak CPU from DPACK (below) so we can accurately size new 14G servers using Skylake CPUs.
Peak CPU 468GHz Net CPU 1193.08Ghz Cores 472
Now clearly 1 GHz on a Skylake CPU with typical workloads (i.e. Hyper-V/VMware running a general range of Windows/Linux servers) might be equivalent to:
- v4 0.8 GHz
- v3 0.7 GHz
- v2 0.6 GHz
- v1 0.5 GHz
Now I know that the performance difference depends on the workload so the above would need to be fairly conservative, but I assume it is possible.
Just having the above rule of thumb numbers that are endorsed by Dell EMC would be a big step forward.
I assume this is going to be an even bigger problem when 'other vendor' CPUs are available at the end of the year, as a GHz of Skylake may be quite different to a GHz of another vendor CPU.
The bottom line is does it make sense to size based on Peak CPU GHz with a weighting/normalisation based on the CPU architecture/generation and if so how do we do it?
--------------------------------------------------------
Does anyone have any opinions on this?
0 -
Rowan,
From what I understand, our server division is building a guidance wizard that might be out as soon as a couple of weeks.
However, this is really something that needs to be answered by Intel. I think it would be hard for a server OEM to make claims here since if they don't make the chip.
I have seen some generational guidance documents, but they all market as non-postable.To be safe, you might look 15% over v4 or (~1.15%), 20% over v3, 30% over v2.
Disclaimer: These are very unofficial fragments that I have been able to pry out of reluctant people :)0 -
Sam,
Do you know if the 'guidance wizard' was released?
Cheers,
Sam
0 -
Sorry. What is the guidance wizard? I don't think I am familiar with that.
0 -
I was referring to your earlier post, July 28, you mentioned:
"From what I understand, our server division is building a guidance wizard that might be out as soon as a couple of weeks"
Do you know if it was released?
0
Please sign in to leave a comment.
Comments
5 comments