This product speeds up VM IO speeds dramatically. It doesn't work with ESX yet.
My question is what could they be doing that would be able to generate such large gains in IO (up to 30x)? It talks about caching? I would like to know how this would work with a VM load that I would really be concerned with regarding IO: Databases?
I guess I will chime in, now that both John and David have spoken up. The reality is that with larger applications running inside of VMs, typically there is a chain reaction of performance bottlenecks that take effect. An example would be running a large DB or the like and running into the IO bottleneck. What are the effects?
1.) You have given all of the memory that you can to the VM and the DB or other app has consumed all of it. Why did this happen? Well, with all applications, there is an attempt to load as much data as possible into memory to improve performance. What does this cause? It causes OS swapping to occur and it causes the kernel of the OS (all of the OS references right now are Guest Related) to grow as IOs are pending.
2.) So now all of the memory is gone, the swap file is being used in place of memory, so what happens? More disk IO! This causes what? The kernel to bloat more dealing with more outstanding IO requests.
3.) The application starts a downward spiral in performance until it either isn't very useful or it finally catches up on servicing the requests and then awaits being bogged down again.
4.) All while the CPU is trying to handle the kernel overhead, the IOs themselves, swapping, and the context switching and thrashing taking place.
If you have a very performant IO subsystem in place, in theory you shouldn't run into this type of bottleneck. The issue is that in almost all scenarios out there today using virtualization, you will run into something like this. Which is why we created DXtreme.
The simplest way to look at the above scenario is when you are using a laptop/desktop computer and it starts to run slow. What does the average joe, computer person say to fix this type of performance issue? Add RAM (memory), this is because memory lessens the dependency on the hard disk and therefore increases the performance when the system is IO bound. It will also decrease the swap file usage, again decreasing the disk IO.
I hope that is somewhat insightful.
I was being a bit predantic when I said that the Hypervisor adds 25% it was an easy number to reel off!
The benefits of any overhead to me is far out weighed but not to my application teams!
Onto your statements I understand what your product does, with a product like SAP that runs on SQL it pages LOADS purely because it sucks memory with individual instances, this also is the case for SQL as it grabs all the memory.
Im probably going to have a shot with this on Vmware server but I run ESX so would rather wait until this is available for ESX.