DELL'S M905 POWER EDGE blade server hits the shops today meaning we have got a press release claiming all sorts of things from AMD and Dell.
Firstly the pair say they have the only 16-core x86 server to support 11 tiles or 66 Virtual Machines if you are using the VMmark performance testing.
The new server is now available along with the Dell PowerEdge M805 blade. Both servers are uniquely designed to use AMD-Virtualisation technology.
Patrick Patla, general manager, Server and Workstation Business, AMD said that the quad core Opterons under the bonnet of Dell’s servers will help speed mainstream virtualisation adoption in the industry.
He was enthusing about AMD's Direct Connect architecture which he said helps enable the efficient memory and I/O capabilities that memory-intensive virtualisation servers require, while alleviating the bottleneck of traditional front-side bus architectures. We all know how painful it is to have a bottleneck in your front side bus... particularly if you have not got any spare lubricant.
The servers also have AMD's Rapid Virtualisation Indexing, which Babychipzilla says can significantly improve virtualisation performance through more efficient virtual machine memory management.
The new Quad-Core AMD Opteron processor-based Dell Power Edge line of servers can support VMware ESXi 3.5, Microsoft Windows Server 2008 Hyper-V and Citrix Xen Server Dell Express Edition integrated hypervisors.
We are expecting another enthusiastic press release from Dell to bring cheer to our Monday morning later today. We guess this will include prices. But it looks like AMD beat it to the punch. µ
16 cores = 66 virtual machines? Leaving aside the rest of the hardware required per virtual machine, since when is less than 1 core (1 CPU) sufficient to manage and deal with a single operating system, which in these cases is supposed to support various server-based applications. 16 cores means at most 16 machines running, probably quite slow, including the host OS. I'm not saying you can't run as many virtuals as can be squeezed on, I'm saying that using less than 1 core of processing power per operating systems running, is going to slow the service down to a near halt. A virtual machine still needs real life hardware to the same specs as any host operating system. If your server board has 16GB RAM, and you have 16 operating systems using that, effectively they have 1GB RAM each. Yes clever time-slicing and the likes can be used to share those resources, but that doesn't change the available hardware specifications. It'll still run slower, & the resource management itself takes up resources. Anyone that actually installs virtuals and runs them & then uses them, knows that. & I hope they have their hard drives ready for heavy defragging, or very optimally partitioned & allocated to begin with.
Anyone familiar with the "v" appliances from 360is (http://360is.com/downloads/360is-V1624.pdf) and how they compare with this product from Dell? Sounds like a similar product at a better price, but new in the market? KBW
It appears that you have no experience with real-world virtualization... I've designed and deployed dozens of them (albeit using exclusively VMware technology) and can say that oversubscribing ram and CPUs is commonplace and an industry-standard thing to do. For example, in one environment we typically have approximately 20-25 virtual servers balanced across 10 ESX hosts (using DRS, HA, VMotion,...) and these hosts have 8 CPU's each so that gives a 2.5:1 ratio. As for ram, the host servers typically have 32 GB each and the guests have 1-4 GB. (BTW - this example have P4 era Xeon 5000 series processors, not the nicer Core2 generation stuff of today) VMware uses various technologies to reduce overall memory requirements by quite a large percentage. For example if you have 15 VM's running all with Windows 2003 Std and 10 NT boxes, it will take all of the common memory-space and reduce it down to one copy. That means you can see VM's using as little as 10MB of actual ram (and the balloon driver piece is gravy on top of that). With regard to CPU's, if you have CPU intensive or latency sensitive applicaitons, keeping a 1:1 or greater CPU ratio is a decent idea but I've seen 20 fairly busy VM's running on a 2 socket box and the end-users were never aware of the lack of a direct hardware based service. Even some of my smaller deployments using VMware Server 1.x running on a Linux host can have 20+ VM's running... Your last statement about storage proves you have no idea about IOPS, spindles, SAN or other Enterprise technologies... Getting back to the topic, good for Dell but I'd still vote for IBM BladeCenter over Dell PowerEdge Blade technology regardless of the premium to do so. (even though Dell does make some competitive mid range gear).