- Channel HP
- :
- Enterprise Business Blogs
- :
- Servers
- :
- Mission Critical Computing Blog
- Mark all as New
- Mark all as Read
- Float this item to the top
- Bookmark
- Subscribe to RSS Feed
- Invite a Friend
Server Power, Cooling, and Density, Tukwila, and more...
Last week, we had a training event for some of our internal as well as partner reps out here in California. One of the topics that I got to present multiple times was Greener UNIX.
I have asked the question, multiple times, about which has a higher importance: increasing server density, or allowing systems to take up more physical room but consuming less power. While I often get a mixed response, with the economy starting to recover, the answer was almost anonymous "help me get more stuff in my datacenter."
Most customers seem to like higher density - but hate the idea that while they might get much more performance, they might only deploy half full racks (preferably with blanking panels in them) since they can only power or cool at a certain density. You get more performance, smaller footprint, and a much better price/performance, performance/watt, and performance/space. Hopefully the half full racks become less of an issue.
I also learned a few things, and got reminders about others:
- Each watt saved on components has a significant multiplier - often >2.5x - by the time you get to the overall power and cooling draw.
- High Efficiency power supplies, such as those used by HP BladeSystem and other HP servers, are great - but may save a few percent of system power usage. Turning on CPU power management, such as Green Active processors in HP-UX, has a much bigger impact on the total power usage than a more efficient power supply.
- At least some of the people who work with customers mentioned that their customers are aware of the power savings features in HP-UX and on HP Integrity servers - even if they haven't implemented them yet. Implementation would be better, but at least people are talking about it.
Needless to say, there was talk around the Intel Itanium 9300 (aka Tukwila) announcement. I've seen some comments on the blog that people want to know about which servers will include these processors. The one comment, from the Intel press release, is "OEM systems are expected to ship within 90 days."
And on a personal note, I'm definitely enjoying the Olympics. However, having been born and raised in Canada, I'm happy to see the Canadian men's hockey team starting to have more success. The games against the Swiss and Team USA, my current home, were stressful to me as a fan. However, after dispatching the Russian team yesterday, I'm hoping to see a Canada and USA rematch - in the gold medal game.
Jacob
-
BCS
-
c-Class
-
Green Active
-
HP BladeSystem
-
HP Integrity
-
Intel
-
Tukwila
HP-UX 11i v3 and Power Management
Recently, we were running to power and performance benchmarks on a server in our lab. We were using HP-UX 11i v3, update 5 and we were testing the impact of some of the power management features on the system.
The power management features that are available as of the HP-UX 11i v3 Update 4 release (0903 release) include Green Idle processors (pwr_idle_ctl), Green Active Processors (pstatectl), and more (see http://h71028.www7.hp.com/enterprise/w1/en/os/hpux11i-power-cooling-overview.html).
Green Idle processors basically takes the idle processors and drops it into a low power idle state, also known as the c-1 state. This is the same idle state that unutilized Instant Capacity processors are placed into automatically by the firmware. One interesting note is that when the OS places today's Itanium Montvale processor into C-1 state, it wakes up the processor every 10 milliseconds to handle a timing interrupt (if I remember correctly... I know the 10 ms is correct, but it could be a different interrupt). Anyway, basically, it wakes up the processor every 10 ms, at which point, it checks to see if there is any queue for it process, and if not, drops back into a lower power idle mode.
Green Active processors takes working processors and drops them into lower power p-states, basically slowing the frequency of the active processors. While the p-state can by set manually (to high performance or low power modes), the most balanced mode is the dynamic mode. This basically changes the p-State based on the activity level of the processor. If utilization is low, it drops it into a lower power mode. If it is in the 50% range, it increases the frequency, and around 80% utilization, it puts the processor in high power mode (P0 state).
So, why the explanation? Well, we were testing a system, and with Green Active processors and Green Idle processors enabled, it's maximum performance was within ~1.2% of our base run (and since we only ran it once for each run, it could be within the run to run variation). But here is the big difference: with both Green Active and Green Idle processors enabled, power consumption at 100% utilization dropped by about 8.5% and power consumption at idle dropped by about 11.7%
Now, I know that people who purchase mission critical systems want to get the maximum performance from those systems. However, would they be willing to activate these power controls since it has a minimal impact on performance? Even more importantly: we ship systems in maximum performance mode today, and those people who install the HP-UX 11i v3 updates that have these features can rest assured that the default is the same high performance mode that their previous releases that didn't have these commands used.
My question: have you enabled these power management features on your systems today? If not, why not? Should we change our maximum performance defaults, and ship our systems with power management enabled, and turn on dynamic power management with HP-UX 11i v3 future updates? Comments and feedback would be greatly appreciated.
Oh, and if you are interested in learning more? Check out the web site listed above, or the white paper at http://h20195.www2.hp.com/V2/GetDocument.aspx?docname=4AA2-5482ENW&cc=us&lc=en.





