Idle RHEV Hypervisors save power?

Quick datacenter power consumption quiz:

Q: How much power does an idle server consume?

A: Way more than a powered-off server.

VMware vSphere saves energy — and money — in your datacenter by powering off unused hosts during off-peak periods.  Take a look at this new technical paper on VMware DPM that covers the technology in-depth.

Despite competitor claims, this powerful capability is unique to VMware vSphere.

Imitation: The highest form of flattery

Red Hat Enterprise Virtualization has a seemingly-similar feature called Power Saver that also consolidates virtual machines onto fewer hosts.  However, the RHEV solution stops there — hypervisor hosts simply sit idle with no running VMs; they are not powered off.  No word yet on what the moon and stars in this diagram have to do with an idle host:

Red Hat claims that an idle host requires just 10-15% of the power required for a host with running VMs.  That sounds very generous if you ask me.  Frankly, from the looks of my HP C7000 BladeSystem Onboard Administrator power summaries, it seems quite false.

Update: Bob Plankers actually measured the power consumption of a Dell server in a few different scenarios and found that idle systems still draw significant amps.

Red Hat claims their new virtualization product is “best in class.”  Really?  Which class?

Tags: ,

20 comments

  1. Scott Lewis’s avatar

    For sure a completely idle server consumers less power than one running at 100%… but this is just plain absurd! That anybody would call this best in class compared to DRS/DPM is just flat out lying.

  2. Scott Vargo’s avatar

    Good points but if your so concerned about powering down VMs at off peak times then your capacity management and optimization planning is crap. Consolidation and optimization are the key I want my machines running 60-70% utilization everyday all day, now that’s true return on investment especial if I’m going to have to pay those excesive prices VMware charges for their product. Now on the flip side if I’m not going to do C&O then almost as good as VMware is probably the way to go since your probably not utilizing all the features vSphere has to offer then go with Xen or KVM and save the money.

    1. Scott Lewis’s avatar

      Sorry my business doesn’t run the way yours does, but nasty comments aside (my planning is crap, what do you base that on?) – the fact remains that OUR business has PEAK loads on CPU during the day that are 8-10 times (on certain days of the week specifically) higher than a ten hour period in the middle of the night. Sorry, that’s a bit tougher to balance, and no, we can’t shift the work to another part of the day, it doesn’t fit what our employees do. On the flip side, at night, we chew up more disk I/O, so powering down hosts during this time makes perfect sense.

      A few contracts back, I worked for a health products seller than did 60% of their annual business in a two month window where they had a rather large discount sale. People would buy in bulk once a year, and then crawl back under a rock. This was before DPM, and this app wasn’t fully virtualized, but there’s another market. Although granted this was so predictable, it could be managed by hand, as well.

      Eric’s points remain valid though… if you are watching power, and don’t have 24×7 constant load on your boxes, there’s no question that DPM seems to trump any other vendor’s options right now.

      Is *THAT* a reason to buy Enterprise alone? Probably not, despite how high my power bills have gotten, most of my utilization is in remote data centers that don’t do metered billing – but it’s certainly ONE more thing. You see, I don’t buy expensive VMware licenses because of one feature, but rather because of 10. HA, Fault Tolerance, DRS, DPM, Storage vMotion, better integration to EMC’s management tools, etc, etc – all together, it was a big win, and well worth the investment.

      Do I hope the other products catch up a bit more? Of course. It will drive new innovation, or perhaps drive the cost down, but for now – things are so lopsided in feature sets and market share, that it remains a tad expensive.

      Doesn’t make what we do crap, however, any more than you using a competing product doesn’t make you “cheap”. Perhaps your business requirements differ than mine. i suspect so, in fact. But don’t think you can sway me, I’ve been attacked in comments for years on similar blogs. It’s worst on Slashdot, actually, but basically, anytime you give a glimpse into your environment and what products you use, somebody comes out of the woodwork and assumes you have NO idea what you’re doing, you don’t plan, and your environment stinks. If you can assess a network with the information given above, I submit that your planning skills might also be questioned.

      1. ScottV’s avatar

        Not attaching but I can tell you that I can save you more money with a solid consolidation/optimization plan then with powering down VM’s. However I do agree it’s the features that sell the product. Heck for what I do we just turn over the extra Cores to the HPC guys when there not in-use doesn’t fit everyones model works for us. Same with VMware simply put the hypervisor battle is over the game has moved to the desktop again and that’s were someone like Citrix (albeit they have their own pricing issues) that own’s that space has the edge. Regardless VMware, Citrix, Microsoft are all fighting a street battle while the rest of us could care less we are headed to IaaS and PaaS environments so we really don’t care it’s just rental space to us.

        1. ScottV’s avatar

          Oh BTW notice this isn’t about RHEL or KVM since they really don’t matter anymore. Can’t argue about something that can’t deliver.

        2. Eric Gray’s avatar

          Note that power management has nothing to do with powering down VMs — the hosts are shut down to save power.

        3. Wade Dicks’s avatar

          I have to agree with Scott. I have a client that’s a 24/7 operation, but between 8 to 5 they have 4 times the load as the rest of the day. So DRS/DPM is perfect.

        4. Collin C. MacMillan’s avatar

          turn-around for a DPM server is in 5-10 minutes – potentially more depending on workloads and boot times. I can see a use case when load shifts every 20 or 30 minutes like batch loads, but otherwise (aka. the general case) DRS+DPM seems like the best in class solution.

          I use DPM a lot and it saves a lot of power – especially when you take cooling into account. It would be interesting to categorize the load profile of a host migrating it’s workload to determine how long it would need to remain idle/off to make up for the artificially increased load of the migrations.

          All of my DPM hosts are diskless or boot-from-flash ESXi systems/blades so mechanical stresses of the power cycle are minimized. It would be interesting to contrast idle server waste energy vs decreased system longevity in standard server configs. Maybe that would show an additional benifit to RHEL’s approach?

          What I’d like to see is a sleep mode in DPM. It taunts me with that “sleeping” icon, but I know it’s off. I’d like the option to halt the system without a total power down – allowing for a quick restore of resources. Then, an aggressive saving mode could halt the system for 5 minute periods without the fear of resource saturation during a lengthy wake-up…

          As for good consolidation planning, I for one think DPM is a great talking point for that stage. The incremental cost of the additional node(s) is nothing compared to the additional agility in infrastructure it buys. With DPM, those resources can be brought to bear during the enterprises most demanding times and present a zero footprint the rest of the time.

          VMware might even be able to license DPM tech based on node uptime instead of per CPU – maybe CPU hours instead – to push the value prop. Ex. You’d buy 8 CPU hours of Enterprise and get to run 12 CPUs for half a day and 4 CPUs for the other half. That’s what I call vFlextime 🙂 With an smart scheduler and balanced node configuration, the scheduler could insure that average runtime of each node was consistent over time – powering down the node with the highest uptime counters first, etc.

        5. Adam’s avatar

          SOOOO basically you work for VMWare and started a blog to complain about your competitor to no end. Have you taken into account for the fact that RHEV is a fairly new product and is continuing to improve on a regular basis? Yes, the feature set is not the same as VMWare, but it is very close and within a few years will probably be nearly identical. Why don’t you post some numbers on here related to the cost differences between RHEV and VMWare?

          Redhat only purchased their virtualization software within the last few years and they are already getting their manager off of Windows. How long has VMWare been around and how long has it taken them to do the same?

          Its been a while since I’ve seen such a narrow-minded and biased blog.

          1. Eric Gray’s avatar

            Adam,

            Yes, I have taken into account that RHEV is a new product and the feature set is not the same as VMware.

            Unfortunately, Red Hat has not taken that into account. They are actively promoting their product as equivalent to VMware vSphere. I think we all agree that it is not.

            If you can afford to wait a “few years” that is fine. Most cannot.

            And yes, my writing is biased toward VMware. I do not pretend to be an independent third-party (they don’t exist, by the way). But everything I write is factual.

            Thanks for reading,
            Eric

          2. Scott Lewis’s avatar

            Yes Adam, he works for VMWare and started a pro-VMWare blog. Are you new to the scene? I read a lot of virtualization and storage blogs. Some third party, many by employees. It’s nice to find a good mix of people you agree with and disagree with, in the same vein that I enjoy hearing democrats and republicans on talk radio, despite being just one of those two myself. 🙂

            Your retort stopped me at two places:

            1) The feature set is not the same as VMWare. You nailed it here. It’s not. For those in which the feature set encompasses EVERYTHING they NEED and WANT, they should look at it. For those relying on (or planning on relying on) some of the features that aren’t there… well, it might not be the product for them. This is not an industry where one product needs to serve everybody. But until some of the newer guys catch up more, and leap frog in other areas by introducing new features unavailable elsewhere – well, VMWare will continue to be the marketshare leader. That’s not surprising. It’s like the mobile phone space, every month there’s a new “iPhone Killer” that gets marginal (many) to good (some) reviews, but never quite knocks the iPhone off the pedestal. Once the reviews hits they start saying… (moving on to #2 to finish that thought)

            2) Redhat only entered the market recently and …

            Doesn’t matter when they hit the market, or what they plan on doing. I’m getting ready to buy another 8-10 licenses for a project. I need something in 2010, so if RHEV can’t meet all of my requirements, I won’t buy it. Back to my point from above, when the reviews hit, everybody always likes to throw in the “well the iPhone had a 3 year head start, and didn’t even have cut and paste for a year and is just now getting background… wait till you see what WE do in a few years”.

            The problem is, I don’t wait. I need a phone now. I need a hypervisor now. I own Office because I need certain things now, and OpenOffice is close… but not close enough. The problem with being late to the party is you still have to win. If I start a 10k 10 minutes before you, and the objective is to win, well… you have to beat me. Granted, if you’re a GREAT runner you might actually do that, but that’s besides the point. If I finish ahead of you, you can’t say “well, my stride looked better, and if I started when you did, I WOULD HAVE BEEN AHEAD” because it doesn’t matter.

            PS: Would I like the option of a non-Windows based management suite? Sure. I don’t really care if the server end is Windows based, but a Mac client would be nice. Is it a deal breaker, no. Who is buying enterprise virtualization products and has engineers without access to Windows (or at least without an RDP client to hit a box and then open the client)?

            1. Eric Gray’s avatar

              Well said, Scott. Thanks for that contribution — great metaphors.

            2. Doug Wandell’s avatar

              The foundation of this article is completely wrong.

              RHEV has had the capability to power off nodes in a virt cluster since the beginning. You can define a power management strategy for each cluster. When a hypervisor host is not needed, RHEVM will fence a node or nodes, migrating VMs as neccessary. The actual fencing happens via Python (Expect) scripts that access out-of-band management cards in the hypervisor nodes.

              A variety of out of band management cards are supported – DRAC5 worked for us until we migrated the stack to Dell PE R610s, and there is (at last check) no fence scripts that can speak iDRAC6. So we’re out of luck for now, but the point of this is – what you’re describing is completely wrong.

              We have a handful of folks here where I work (5000+ employee govt. and defence contractor) who get very religious about VMWare, so I am used to your mindset, but please don’t spread inaccurate information.

              1. Eric Gray’s avatar

                Doug,

                Thanks for taking the time to comment. I do not understand how you can say RHEV automatically powers down unneeded hosts. Yes, there is a manual button to start and stop hosts, but from what I experienced in my own hands-on investigation is consistent with the product documentation: hosts remain powered on, but idle.

                In fact, Red Hat’s datasheet on Power Saver says:

                The host will now run idle, consuming significantly less resources – typically 10 to 15% of the power of an active server. If other hosts in the cluster reach the maximum service level the System Scheduler will live migrate virtual machines onto the host and re-balance workload within the cluster.

                And the demo here also confirms that hosts are idle — not powered off.

                Is there an undocumented configuration option to enable power on/off of servers?

                Eric

              2. Doug Wandell’s avatar

                As it turns out, our solution does involve a little grow-your-own scripting, as confirmed by members of our team. Apologies for my mistake – this is not an out of the box feature. Apparently this was reasonably easy to implement and while not as sophisticated as DPM, works well. Actually makes you wonder, why support auto-migration power saving without auto-fence?

                Anyway, works for us, but not without a bit of hacking. Thanks for the info.

              3. Abigail’s avatar

                go rhev

                1. Walt’s avatar

                  I couldn’t help but notice nobody compared the price of VMWare to RHEV. Let’s face it boys and girls, virtualization is now a commodity and VMWare is priced like they have the only solution. When I took the RHEV cert class last summer they talked about live migrating and spinning down hypervisors off hours. What a difference a couple of months make. Oh, live migration on VMWare, ka ching, something else you have to buy.

                  1. Scott G. Lewis’s avatar

                    Walt,

                    Price didn’t factor in much a year ago when this article was posted. As you’ll see from the comments, it was a split between people relying on features that one product didn’t support, and people who didn’t need those features and bought cheaper. Has RHEV caught up in the past year? Perhaps. I’ve moved on to a new firm, who is XEN and VMware based, so it’s not really something I’m currently tracking.

                    1. Walt’s avatar

                      I did notice it was an ancient (in technology time) thread but I happened across it late. I honestly don’t know how well RHEV is catching on. I would think RH has the gravitational mass to do well. I like RHEV but, when I did the cert course, it required Windows to host RHEVM. Hokey. I think my underlying assumption, independent of RH, is where now-a-days any nearly current release of Linux comes with KVM which has all the live migration and facilities of virtualization and it’s free. So, in my opinion, with RHEV you are paying for a better UI (than virt-manager) and a phone number to call if something goes bump. It’s hard to beat free or nearly free. So I’d think VMWare will be under pressure to reduce margins to compete. We use VMWare at work and I use VMWare Workstation at home along side my UEC. It’ll be interesting to see what RH has to say during Summit, next week.

Comments are now closed.