Tell me more ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.
up vote 3 down vote favorite

I am working on pricing out 4 Dell R710s for a new Hyper-V cluster. I have a choice between the following two configurations, for the same pricing:

128GB of 800Mhz RAM per node

Or

96GB of 1333Mhz RAM per node.

CPUs are dual Xeon E5645.

It would be more ideal if I could go with the 128GB option, but I don't want it to come at a huge performance cost...

In this type of a situation, would the difference in the RAM clock speed produce noticeable performance gains or losses in a typical VM environment?

I could get away with the 96GB for at least 2 years.... So is the performance boost worth it?

share|improve this question
3  
never seen a hyper-v enviro impacted by mem speed its almost always io thats the problem! – tony roth Apr 1 '11 at 5:01

3 Answers

up vote 7 down vote accepted
+25

I would go with 96GB, not for the reasons stated, but due to the limitations and configuration options of the R710 server. For best performance, you will want to use what Dell calls "Optimized Mode" for your memory and fill all three channels per processor. By utilizing all three channels, each memory controller can be independently processing different requests.

However, if you fill all three DIMMs in a channel, the bus speed will automatically be clocked down to 800Mhz. In order to keep a faster bus speed, you can only fill two DIMMs per channel, and even in that configuration, the memory will be clocked down to 1066Mhz. 1333Mhz is only available with a single DIMM on each channel which would limit total capacity to 48GB.

With this in mind, I would install 12x8GB 1066Mhz dual-ranked DIMMs, placing 2 DIMMs in each channel on each processor. The DIMM slots used would be A1 through A6 on processor 1, and B1 through B6 on processor 2.

The last benefit of this configuration is that you could easily add another 48GB to the server down the road.

Sources:
Dell 11g Memory Whitepaper
Dell PowerEdge R710 Owners Manual

share|improve this answer
additionally, if you KNOW that 96 Gb will last you 2 years, why get slower ram that it's likely you won't use. It sounds more likely that you would add a newer server before you needed more ram. – Jim B Apr 11 '11 at 15:01
up vote 4 down vote

The 128GB will not only be slower per-stick, it will also not be tri-channel configuration. The faster RAM + tri-channel will probably make a noticeable difference.

That said, you have to size based on your VM load. If you will not have enough free memory to run all VMs on 3 hosts (in case of a host failure), you should go with the 128GB. Otherwise, go for the speed!

share|improve this answer
Do you have any resources that you can point me to that describes tri-channel config, etc (?) – Josh Brower Apr 1 '11 at 11:33
Each "channel" has direct access to the CPU core's I believe, so it allows more RAM to be used simultaneously during heavy memory I/O. Honestly unless you expect a LOT of users hitting this boxes and high CPU utilization I wouldn't expect a difference. My typical Hyper-V box has 8-16 core's and even with dozens of machines in production CPU sits at 25% or less. The two reasons I have to add another box is maxed out RAM or disk I/O like Tony said. It just depends on if fitting more machines or faster machines is your priority, but I doubt "faster" will be noticeable to avg user experience. – Bret Fisher Apr 1 '11 at 16:03
1  
tech.icrontic.com/articles/what-is-tri-channel-memory has a nice little write up. Here are memory bandwidth stats: en.wikipedia.org/wiki/… - 102.4 Gbit/s for 800Mhz/dual channel; 256.0 Gbit/s for 1333Mhz/triple channel. This may or may not make a significant difference depending on the workload, and it just one of many factors in performance. – Paul Doom Apr 2 '11 at 1:24
+1 As others have noted, Hyper-V boxes are almost always bound by RAM Size and IO (including "disk" IO and RAM IO). That noted, unless your VMs have large memory requirements, putting 20 to 40 VMs on a host commonly runs up against IO; it might be more cost and performance effective to get 6 servers with 64GB RAM (just fishing as I know nothing of your environment or workloads). – Chris S♦ Apr 7 '11 at 12:51
@Bret's comment: Just noting that an idle CPU doesn't prove that RAM speed isn't a bottleneck, it might actually support the opposite conclusion. Pretty sure that if the CPU is waiting on memory access, it's not actually doing work from a perf counter perspective? (anything the CPU waits on will generally cause it to idle) – TristanK Apr 11 '11 at 9:52
up vote 1 down vote

I agree with some of the previous posters. The biggest question you need to ask is if you are going to run low on RAM. Also using more than %80 of physical RAM for normal use is dicey. Of course peak usage is one thing, but while faster RAM is better if you're not going to have enough that is a big issue.

Also where is your bottleneck going to be? Are you using a SAN or DAS? If you max out your storage arrays IO then the faster RAM is not going to do you that much good, and the extra RAM will matter more.

If you aren't going to max out your RAM and you're running an enterprise SAN (LSI, EMC etc) then the faster RAM would be a better bet.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.