I was recently reading a blog by David Hitz and he raised an interesting question about where we were in regards to server virtualization:
Something has been bugging me about the market share numbers for server virtualization. Is the trend is just getting started, or is it almost finished? The numbers I’ve seen say that under 10% of all X86 servers have been virtualized – maybe 7-8%. By that measure, the trend of converting physical servers into virtual ones seems to be quite early.
I’ve got to say, I agree, but with a caveat. It’s actually not the beginning of the trend in general, it’s the beginning of the second wave. When you look at systems holistically, when you get into enterprise applications or services, the applications far exceed the scope of a single physical machine. When you’re looking at pools of externalized storage married with blade servers and virtual machines, what we really have are new, open, mainframes.
Mainframes, reborn:
If we look at a description of why people are still using mainframes, from an article over at serverwatch.com, we see a description of their advantages that sound eerily familiar:
Logical partitions, known as LPARs, can be used to run multiple operating systems, including the z/OS, z/OS.e, OS/390, Linux on zSeries, z/VM, TPF, VSE/ESA, zVSE and zTPF. All major databases and enterprise transaction processing environments run on the new mainframes, including CICS, IMS, WebSphere Application Server, DB2, and Oracle.
Sounds sorta like virtualization, doesn’t it? You centralize and thin provision, picking up resources for where you need them by using the non-utilized resources of other servers. In addition, you don’t need to have two of every single server — we’ve effectively been doing raid 1 for mission critical servers, having a “db1” and “db2” for those mission critical database servers – now, instead, if a blade dies, the server gets transparently moved to another. This was the same idea that drove the adoption of the mainframe. You just need to have enough extra capacity for the physical needs – cpu, ram, disk – not for the logical – “second webserver”, “second database server”- it’s also the same argument for san and nas if you thik about it.
A happy accident for netapp:
I meet a lot of netapp customers, and in fact, netapp is a big driver of the virtualization trend among their customers. They way they do flex cloning and read caching was a *very* happy accident in regards to virtualization on their devices, the performance is amazing; if you put your vms on a flexclone or use their deduplication stuff it ends up having incredibly high cache hit rates. Also, the blocks are all pointing at the same places so the disk utilization is small. It’s cool stuff they’re doing there.
The swing of the pendulum:
As for the overall trend, I think we’ll see a move towards more and more virtualization. Eventually, we’ll hit a point where we end up centralizing and vms are the norm and we’ll start creeping back in the opposite direction. Single server computing resources by that time will be increadible and we’ll see the pedulum swing in the opposite direction. For right now though, I think we’re going to see a rebirth of the mainframe, under the alias of virtualization, and as for me, I’m a pretty big fan of that idea.