Today VMware announced that performance gains of nearly 50 percent are achievable by dynamically balancing running virtual machines according to resource demand with VMware DRS.
The VMware Performance Team has repeatedly demonstrated how vSphere is the superior platform for your production workloads. Check out the latest article on the VROOM! blog for all the details.
By running substantially more virtual machines per hypervisor host, VMware provides the lowest cost per application — even less expensive than so-called “free” virtualization solutions.
You might be tempted to compare this tremendous success with first-generation offerings from other virtualization vendors, such as Citrix WLB or SCVMM PRO Tips. Coincidentally, both of those other solutions are not integrated –the resource balancing is performed by an external system that is monitoring virtual machine performance. DRS is part of VMware vCenter and purpose-built for the virtual environment.
Note: the version of System Center Virtual Machine Manager 2008 that supports live migration is not slated to be available until late December 2009. That means Microsoft virtualization environments cannot yet even attempt to accomplish what VMware does today. And by today, I mean June 2006.
Meet the Engineer
In this video you can hear one of the main engineers behind VMware DRS:
I am about to build a DRS cluster for testing. One thing I am a bit confused about is whether VM’s can use resources from more than one host at a time.
Example: If I have a SQL box that gets max’d out and needs more RAM, can it use RAM from another host in the cluster? Or does is it constrained to finding a single, larger host?
That’s a great question, and I hope your testing goes well.
Virtual machines cannot use resources from multiple hosts simultaneously, so your SQL Server would need to migrate to another host with more RAM.
However, DRS could still help you in this case by shuffling the VMs around in such a way that frees up the most amount of RAM for your DB.
Great post, thank you. It is great to see DRS – the first practical automatic load balancer.
Do you know what are the real scalability limits of DRS today – I have yet to see any mentions of beyond 6-8 physical servers…
Also to the best of my understanding DRS does not take into account the disk i/o channel loads – is this true ? It would explain the huge cost of the storage used in the test.
My back-of-the-envelope calculations for the test above show that the storage is about 20-25 times (yes, times) more expensive than the servers (more details at blogs.virsto.com).
Wouldn’t it make more sense to save on storage ?
I have used DRS in larger clusters than that. In one environment for example, VC handled DRS for 4, 6, 12, and 16-node clusters in the same virtual datacenter. I never knew DRS to have any problems.
Jonathan, thanks for your input here. As you say, DRS works great with clusters larger than 3-4 nodes. I suspect folks doing tests and experiments just want to create a sample environment that is fully functional while minimizing the logistical burden, thus 3-4 nodes.
BTW, I was debating how to handle Alex’s comment — it’s obviously plugging his startup. I decided to let it stand.