I have lost touch with the hardware side of servers recently more so in relation to virtual performance, but ive recently been looking to replace existing servers, currently running BL60c G5's with the 4 x E7440 which gives me 16 cores @ 2.4GHz 16MB L2 per socket, now as i understand it these are built on the core2 architecture.
The servers that are looking to replace these are the BL620c G7 running 2 x X6550 which will give me 16 cores @ 2.0GHz 18MB L2 per socket now i know these are from the nehalem cpu family.
I know the nehalem family is allot faster clock for clock but what got me thinking is that I would be dropping 6.4GHz of cpu cycles per second per host and dropping nearly half the L2 cache per core is it that much faster clock for clock that it wouldn't negatively impact the virtual environment?, how would this ultimately effect the performance and consolidation of virtual machines?
I have vmmark for both but the BL680c was dont with verion 1 and the BL620c with version 2 so its not really comparable.
what are peoples ideas on this with the cores increasing but the clock speeds slowing does it have any negative impacts? or are the technology improvements out way the drop in clock speeds?
Clock slowdown is usual in the Intel cycle on each new generation:
But usually new features (more cores, more cache, new VT functions) can beat the minor frequency.
And, if you want you can choose the second generation of Nehalem where you can have ALSO high clock.
CPUs have never really been the bottle neck for me in the past. In my opinion, it's about disk spindles (IOPs) and then physical RAM. If a VM is struggling for CPU, adding more vCPUs generally does help so the more cores you have, the more vCPUs you can leverage. I was also told that VMware recently rewrote there scheduling stack to help ready times in vSphere 4x +. In fact I don't worry about CPUs as much as disk. It also depends on what you intend to run in the environment. If it is CPU you need, then you need to consider your CPU cache which helps improve the read/write of pages in physical RAM for instance. Latest CPUs have many different caching functionalities to help with data and execution transfers which is more important than clock speed. Clock speeds have stayed the same for a few years now as vendors really have broken the barriers in terms of CPU speeds to the point it's not worth invetsing any more in this area but more in the intelligence of how CPUs work within virtual enivronments, like Intels VT..
I was recently speaking to a major service provider at a virtualisation conference and they were saying that they have dropped back their physical servers to 1 physical socket, each with 8 cores simply to reduce licensing costs. I'd say that overall you won't be let down with the CPU clock speeds offered by todays vendors. Not sure this really helps you but I wouldn't over analyse CPU clocks speeds... Hope this helps.
If you need more technical information, there is plenty on the web. A sales vendor will also help but like it's always good to get community real life feedback so your approach is good.
Just be careful of CPU cosmic bit flips. Now there's a thought