Reporting on CPU - interpreting and using for new solution
Have a two part question.
First, customer ran dPACK. Appears to have a mix of physical and virtual servers. Wondering why the speed on all the processors is showing the same? I can determine that Net Clock Speed = CPU Cores x CPU Clock Speed. Looking for a bit more definition on the relationships in this formula, and how it relates to the environment.
Second, determining a new configuration to meet/exceed the current one. I believe this has changed over the years, so just want to clarify where we are now? Given the numbers, do I just take the total Cores count and divide by an appropriate current processor core count? And make sure I can achieve the total Net CPU speed. Using the same customer in the above point, they show 91x cores, Net CPU is 245.7 GHz.
Great questions, Michael. The CPU information reported in a DPACK project can be very valuable in evaluating a group of servers. The way DPACK collects CPU information depends on the Operating System of the server (Windows, Linux, VMWare/vCenter) and in the case of Windows servers, whether DPACK is running on the server or is connected remotely via WMI. Each servers processor speed is unique to that individual server and is collected via an independent data call from DPACK. Your experience of seeing the same clock speed can happen in similar servers (virtual or physical), but is not an indication that all clock speeds are listed the same for all servers in a DPACK project.
It might be of some benefit to use the Publish function within a specific project where you are seeing these speed settings the same for all servers and post it on here for us to review. Please note, the Publish function will anonymize all of the server and disk names to protect the information of customers.
As for your second question on how to use the Net CPU Cycles information when it comes to consolidating and/or replacing servers, you would take the total Net CPU Cycles listed in your DPACK project and multiply it by the peak CPU usage listed in the DPACK project (listed under More Information -> Performance Information).
This would give you the total amount of processing resources used within the servers of the DPACK project. From there, you would determine the total CPU processing cycles for the replacement server (Processor Count x Cores/processor x CPU core speed) and divide that into value you calculated from the DPACK project. That should give you a good idea of how many servers would be needed to replace the current servers, based on CPU usage alone.
By the way, Congratulations on being the first post in the newly released DPACK Community Forum!0
Thanks for the feedback!
I started playing around with calculations, and not sure I'm following the process correctly. For reference, here is the public link:
If I am reading everything correctly, total processing resources for existing servers:
245.7 * .12 = 29.484
I configure a dual processor, 6-core with about 2.4GHz speed = 31.2
This would imply I only need 1 server to replace the entire environment?0
Michael, read this and see if this helps at all.
If I'm looking at a report that is showing Peak CPU @ 59Ghz / Net CPU 735.00GHz w/287 cores; would it be safe to say that they are vastly over-provisioned on CPU?
8 % / 2%
In all likeliness, yes. What you want to look for is the yellow line. If they yellow line in the CPU chart jumps up, then you have saturated cores. So even though CPU might be low, this will indicate single threaded processes that are limited by CPU to be seen.
However, many companies are over provisioned in many elements; CPU is one, but IO is probably the biggest over provisioned item in the industry.1
Please sign in to leave a comment.