Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I'm looking at putting together a report on the current state of our virtualisation project that we began in the hopes of getting some iSCSI storage. Part of it, I'd like to have some real world figures of what people are achieving with their own virtualisation projects.

Basically number of hosts per physical box. Some specs on the physical box would help too as well as the work load. Or if someone can point me to some similar "real world" figures. I'm not interested in theoretical maxes.

For reference, we are running Hyper-V in the main setup. We have a few other virtual server installs and will be migrating those into the Hyper-V setup once I get around the HD space issue that prompted this.

share|improve this question

4 Answers 4

up vote 2 down vote accepted

Most of the environments I deal with are VMware ESX but as far as Hyper-V environments are concerned: 2 x Dell 2950's with 2 Quad Core CPU's, 24 gig of RAM, primary storage is a single Dell Equallogic PS5000 array. Runs 12 Guests (2x w2K8 Domain Controllers, 3 Exchange 2007, 1 SQL, 2 Sharepoint, 1 File & Print, 1 ISA, 2 other Web app servers). Exchange has about 100 users, AD about 1500 (most mail is outsourced). Runs at around 20% CPU load typically, spikes to 40%, Memory utilization is high (70%+). Storage is not optimal at the moment but it's not obviously hampering performance. I wouldn't add any more server guests to this without beefing up some of the specs.

share|improve this answer

Based on some cost analysis and performance testing that I participated in a year ago, the sweet spot for price/performance in a large environment is around 10-12 VMs per physical host. Less than that, you're not necessarily saving money versus using a physical box.

Note:

  • This is based on pre-nehalem CPUs and ESX... the extra memory may change things.
  • I'm assuming hundreds or thousands of VMs
  • If you're using mostly Windows VMs, how you license Windows is a huge factor
  • I didn't account for cost avoidance... the ability to do system snapshots for backup/dr, clustering, etc without buying additional software is great & valuable.

Intel IT has published some interesting materials on this topic; I'm covered by an NDA with them and I'm not sure what I'm at liberty to share, but I'd recommended browsing through their "Open Port" and "IT@Intel" sites.

share|improve this answer

our numbers are fairly similar. We have a 4-node ESX3.5 cluster running around 130 hosts across a mix of FC SAN (some SCSI LUNs and some SATA-based) along with "lighter" servers running NFS. We did originally use iScsi storage for our low-end servers, but switched to NFS because we found it easier to work with in disaster situations. (probably depends heavily on your iScsi box though)

share|improve this answer

We are an ESX shop, but the numbers should be similar.

We've got 2 quad chip, quad core servers. 64 Gigs of RAM each. We have 60 VMs running on these two servers. Each server's memory runs at about 55% used, and the CPUs run at about 20% used.

Our storage is provided by an EMC CX4 SAN.

share|improve this answer
    
Out of curiosity, are you using vmotion to move machines back and forth? I'm guessing since each server runs at 55%, you can't load everything onto one machine. How do you schedule downtime for 30 servers? –  Matt Simmons Oct 17 '09 at 16:23
    
All the web server VMs are in groups of three servers serving the same content. I've got DRS configured to always keep two machines on separate hosts. So if a host goes offline it will leave one of the three web servers offline, and if the one that floats between hosts is offline it then starts it. Same goes for App servers, DNS, etc. Each group of VMs is always spread between the hosts so a host outage would only effect part of each group. Things which are a single point of failure Exchange, SQL, etc are in MSCS clusters to provide HA within Windows. –  mrdenny Oct 19 '09 at 4:20
    
It sounds complex, but it isn't all that bad. Downtime for some VMs isn't an issue as no services are ever actually down. I've had to down a VM host a couple of times now (bad ram once and a host OS upgrade the other time) with 0 customer impact. –  mrdenny Oct 19 '09 at 4:22

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.