Hi,

I am building about 240 computers with Dual Intel Xeon E5507 with 24 GB RAM on a Quanta Motherboard. In addition, the computers will be hooked up via 10Gb LAN Ethernet to the existing network.

On each computer, I plan on using CentOS 6.6 to host 4 VMs using KVM. On each VM, it will host software for taking in data from my databases and processing it to return prediction reports and other analytics. I would like to set up a fault tolerant hosting scheme and allow for live migration, image snapshots, rapid VM provisioning, etc... to minimize downtime.

So, my question to you folks is:

Should I use 4 SSDs or 4 HDDs for each of my build.

The SSDs I have in mind would be Samsung 845 DC PRO 400GB vs WD Enterprise SATA 750GB. In both cases, I would not be capacity limited, as my raw data would be stored on my databases and all capacity on these computers I am building would merely be used to host VMs, queued data, and, before being sent to other databases for further storage, VM processed data.

I am open to all suggestions.

In my experience, SSDs are faster for sequential access, SIGNIFICANTLY faster for random access, and are much more reliable than HDDs. If you can use SSDs for your application without breaking the bank, do it.

That said, I'm a little surprised by your parts selection. E5507 is OLD...why on earth is that your go-to? Also, why do you need 4 drives for 4 VMs? If you're not bandwidth or capacity limited, why separate them? If fault tolerance and uptime is your goal, why aren't you using a RAID setup?

Hello,

Thanks for the quick reply.

Yes, I plan on RAIDing the 4 disks. I will probably go with RAID 5 then and use the SSDs. Part of the array will go towards VM hosting while the other part will be partitioned for queued data storage for processing.

Oh and I am getting a boatload of those CPUs for cheap. They should still equal an AMD FX 6300 correct?

I'm not familiar with AMD's offerings - my only experience with AMD was years ago when their "speed" ratings were so ridiculously out of spec that it bordered on laughable...I haven't used them since.

Is RAID 5 necessary? I was under the impression that capacity was not a concern. If that's the case I would lean toward RAID 1 or 10 instead to reduce the load on the CPU. I generally limit parity RAIDs (5, 6, 50, 60) to hardware solutions with dedicated ASICs to handle the parity calculations. Software RAID can handle it, but bandwidth suffers and CPU load increases, sometimes significantly.

1 members found this post helpful.

Hi,

I currently use dual AMD Opteron 2376 processors so I imagine the "new" Intel processors should be somewhat faster.

I was debating using RAID 10 as well. I am aware of the overhead of the parity calculations of RAID 5 and 6, but from what I have seen, the performance hit is only mild on software RAID. Of course, I do use two dedicated RAID controllers on my database servers with write back enabled on RAID 60. Almost no performance hit there!

I'll probably go with RAID 10 then, as I should still have enough capacity for the data and VMs.