up vote 1 down vote favorite
share [g+] share [fb]

my questions is : if i need to deploy 170 virtual machines and i need HA. is it possible to deploy all 170 vistual machines on one ESX host and in case one ESX host is down. all virtual will migrate to another ESX host.

what will be the traffice effect on the network card. e.g all of the users are using all virtual machines. what will be the performance on ESX host. what will be the hardware requirements for ESX host for around 170 virtual machines. or how many ESX host required to accomodate 170 virtual machines.

Thanks

link|improve this question
feedback

3 Answers

up vote 3 down vote accepted

It is possible to do that yes, ESX 4U1 supports 25 virtual CPU contexts per physical core so you'd need at least 8 cores on each server to run 170 VMs on a single server, plus memory of course. That said HA doesn't migrate VM, that's vMotion, it just restarts the VMs in the event of a host failing. The time to restart these VMs varies based on the number and complexity of VM, and speed of shared disk obviously. There's no real impact on the network as they simply restart from the shared disk system - that would obviously go rather hot but you can schedule a VM restart order if that helps.

If you require zero downtime for your VMs then you need the FT feature, which runs a VM on two machines concurrently but currently limits FT VMs to a single vCPU per VM.

I would suggest that you should buy three or more dual-socket Xeon 55xx-series servers, run them as a => three-way cluster with DRS spreading the load, HA dealing with sudden host failure and FT for any SUPER important-but-low-CPU VMs, all connected to an FC SAN or 10GBps iSCSI box.

link|improve this answer
Out of curiosity, in the businesses you've seen the 170 servers on a single server, what are they using the VM servers for? Virtualized web server front ends? (Curious about load that is placed on the VM's and how it affects the physical host in aggregate.) – Bart Silverstrim Mar 8 '10 at 11:17
I work for a big company and we had a massive server virtualisation project that reduced our data centres from several hundred to a dozen, many were just running on/under desktops or corners of offices. They were all P2V'ed then a month or so later we'd pause them and only switch them back on if some moaned, if after 6 months nobody had said anything they were archived and deleted. So many of the VMs were very old (up to 10 years old in so cases) so very low performing anyway (we resource limited them to 110% of their physical hardware specs). – Chopper3 Mar 8 '10 at 16:43
feedback
up vote 0 down vote

170 machines is a lot...I mean a lot...and you don't mention what your server specs are. I think you might have to have a farm of servers.

To have HA for virtual machines, you normally have a large storage SAN on which you deploy the images, then you mount that SAN to multiple servers. The servers would run something like VMWare with VSphere, and they can handle the failure of one or more servers to migrate the machines to another server within seconds (or less).

VMWare has some white papers and information on this. Other companies have similar products.

So can you have HA for a number of VM's? Yes. Can you do it cheaply? Not easily, you'll need a server farm and shared backend storage plus the licensing costs of the software. You can't do it on one server, and your question sounds like you wanted it on one server. I don't know where you could run one server running 170 machines simultaneously without killing it unless you have a server in the back room of the R&D department of the DoD or IBM...

link|improve this answer
there's lots of companies out there running far more than 170 VMs per host - mostly VMWare View boxes obviously but we have a number of HP DL785 G6's running close to 300, low importance, VMs in a production environment. – Chopper3 Mar 3 '10 at 11:20
Okay. I hadn't had to ramp up that many and the OP didn't give any server specs, but he won't have HA on one host still. – Bart Silverstrim Mar 3 '10 at 14:00
feedback
up vote 0 down vote

you can use RHEV that is bare metal installation and same time how much process your looking pre guest and which hardware your looking buy!! if have hardware in place what i suggest use RedHat HPC solutions with RHEV it give more better performance and less in cost.

link|improve this answer
feedback

Your Answer

 
or
required, but never shown
discard