Hi, I was wondering if anyone of you guys have used VMWare ESXi for any purpose whatsoever. I would just like to ask if there is a website I could use to calculate how many virtual servers I could spawn given lets say 2.4 xeon with like 4 gigs ram.. well hard disk is subjective depending to use.. but i would just like to know how many virtual servers i could load in a machine..
Yes, I personally have tested VMWare ESXi on an HP Proliant ML110 G4 (Xeon 3040 1.86, 4GB RAM) and on a custom i7 system (Core i7 920 @ 3.53GHz, 12GB RAM, Asus Rampage II Extreme, 2x HD4870).
By the way, what Xeon are you using? Dual-core? Quad-core? Hex-core? Nehalem?
I don't think there's an online calculator that can help you with capacity planning. Most online calculators related to virtualization (like the ones from MS and Citrix) are concerned about the license costs.
However, what you can do is run estimates.
From my experience (not empirical evidence), the ML110 can handle 3 virtual servers (MS Windows 2003 Server [Terminal Services for 4 TC's], Ubuntu LAMP server, MS Windows 2003 Server [Active Directory]) quite well.
Perhaps what you could do is assess the system requirements of each VM, then multiply it by a safety factor (I typically use 1.25 - there's no agreed-upon rule of thumb) then add them up. If the capacity of your hardware exceeds that total, I guess it's safe.
answered Oct 15 '09 at 03:51
Watch out: physicalization - low-cost, scale-down server is coming next year!
Scale out computing that are targeting for Web2.0 and Web Hosting is taking an interesting turn from virtualization to “physicalization”. Standard servers that are packed w/ server class multi-cores CPUs (Nehalem, Magna Cours, etc…) from Intel and AMD are increasingly “out-of-balanced” for some scale-out applications. CPUs are stalled/idled most of the time waiting for memory and I/O. Virtualization eases server utilization issue but still can incurs 10-50% overhead for I/O intensive applications. Undoubtedly these CPUs are best in class in term of performance/watt and applicable for apps that do not scale-out easily. Data centers are more concerns with work done/$ and work done/joule than just pure performance. Early this year the term “physicalization” is used by some companies (SGI, Dell) to describe disaggregating larger servers into many smaller ones using consumer parts (desktop, laptop, embedded CPUs, etc…) to drive cost and power down. Intel has seen this as a thread on high margin server class CPU and have demo’ed “MicroServer” concept using desktop/consumer CPUs recently. Intel is pushing this design spec SSI standard body by the end of this year.
answered Dec 06 '09 at 20:18