Wednesday, November 25, 2009

Let’s Talk About VMware DPM

I found this on another site, but had to post it.. Ironically, an Engineer and I were just talking about this yesterday.

Today’s topic is going to be about VMware’s Distributed Power Management or otherwise known as VMware DPM. If you have not heard about VMware DPM before, I found an interesting video interview with Anne Holler who works at VMware as a Senior Staff Engineer. I found this interview at Vladan Seget’s ESX Virtualization site here and it is worth checking out. The video is pretty short, only about four minutes in length and gives a pretty good overview of how VMware positions DPM in conjunction with HA and DRS.

In a lot of the environments I have worked in over the years actually had most of the processing power happen at night during the backup process. Actually at one point I had a cluster that was oversubscribed to put it mildly. If you have ever driven a standard transmission then you know about the red line on the tachometer that you should not go over or the engine would blow up. Let’s just say my oversubscribed cluster was riding that line during the day and when the backup schedule kicked off I actually saw CPU% on the individual VMware ESX hosts themselves reach levels of 102%. Just like clockwork ever day when the backup’s kicked off the host would really max out. Pretty cool to watch but I digress. The point is I am not sure I would really think about using VMware DPM in my production environment. Now if I just built a brand new ten node cluster what was designed to cover the growth for x amount of time and I had just gotten started with putting virtual machines on the cluster then sure why not set that up? I would have plenty of resources available that I would feel pretty confident about that but in most places that I have worked or spent any time at there was not really that amount of spare resources and most customers were always trying to cram as much as they can on the clusters.

VMware DPM was actually introduced in Virtual Center 2.5 and VMware ESX 3.x as an “experimental” feature and only had one way of ‘waking’ a server and that was with a magic wake on lan packet. That worked great until you rebooted Virtual Center or restarted the Virtual Center server service for any reason. After the reboot Virtual Center would see the host as down and was not able to bring it back to life. That was one reason why the technology was experimental at the time. Now in vSPhere, VMware DPM actually has three different wake methods it can use to bring the host back up. Along with using wake on lan DPM will also use IPMI wake and ILO wake. One other really cool thing that was added to VMware DPM was the ability to schedule when an ESX host would wake back up. So if you know you were going to have a heavy load at 9am you could schedule the host to wake up at 8am to be pretty for the start of the business day.

So to recap, I am not sure that VMware DPM would be something I would setup and use on a regular basis. I have also worked in certain environments that the Change Control Board would literally lay an egg if they knew that hosts could and would shut down for any reason.

That is my take on this. So what’s your take on VMware DPM and are you using it currently in your environments? I would really like to what you all have to say and think on this? Inquiring minds want to know.

No comments: