Hi,

I'm writing a MySQL success story of a busy website with big database running on Pegasos.
I would like to add a whitepaper showing some performace / watt numbers for the Pegasos.

Can someone please provide me with the typical power consumption of the ODW?


Many thanks in advance

Gunnar

Well, I have no exact data about the pegasos 2, but in the SOCAL show in LA in 2003, we measured 4 running Pegasos 1 machines complete with flat panel, and total dissipation was 180W for the whole booth (this included the router too though).

So, i believe that a pegasos 1 with a G3 at 600 MHz was disspiating aroudn 15-20 W, (monitors usually come at around 25 to 30W, well my 15" sony one says 27W for example). The 750CXe@600Mhz is rated at 3-6W, i think, and the 7447@1GHz is 9-12W i am told, so this would put power consumption to around 20-25W for a Pegasos 2 with one IDE disk, a graphic card and 256MB of ram.

I hope this helps some,

Friendly,

Sven Luther

Hi Sven,
This is interesting.

I ran some simple MySQL benchmarks using Sysbench 1M records (http://sysbench.sf.net)

The Pegasos 2 with 7447 (1Ghz) reaches 150 TPS (transactions per second).
While an Itanium 2 ( 1Ghz) reaches 180 TPS.
This is a clear performance/watt and performance/price win for the Pegasos.

The benchmark with my setup it quite CPU intensive.
Because of this the raw performance crown is held by Opteron and XEON.
It will be interesting to see how well a 7448 can compete with them.

Cheers

Was this with Konstantinos' fledgling AltiVec hashing code or without?

--
Matt

Without, I haven't merged back my mysql code yet :-/

Konstantinos

>> Was this with Konstantinos' fledgling AltiVec hashing code or without?

I did the benchmark without any altivec optimisation.

But a noticeable part of the benchmark is fetching and sorting 100 strings of 120 byte length each.
Sorting these strings creates on overheat of roughly 10-15% of the whole test.
I would guess that Altivec can give a noticeable speed improve here.


Cheers
Gunnar

Probably next week. There is some equipment to measure it by a local ISP, and I'm also interested in this data...

It's actually next month, but I got the chance to test it only now...

Here is my hardware:
Pegasos2 mother board
1Ghz G4
40G + 80G HDD
CD-RW/DVD-ROM
PS2 mouse and keyboard
ATI 9200 video
512M RAM

I used an EKM 265 digital power meter. The results:

without any load: 50W
with mysql-bench (100% cpu, and lot's of HDD activity): 56W

In comparison, I have an old P4 machine
MSI mother board (I don't know the type any more)
2.4Ghz P4
nVidia Ti 4200 video
and the rest is the same

without any load: 93W
with mysql-bench: 135W

I can use it for another few days, so I can test other test situations, if you are interested.

CzP

Is there any way you could test this with only one HD connected (the 80GB) so that the test system reflects the configuration of the ODW?

Just curious..

--
Matt Sealey

Yes, but it's not worth trying it. Yesterday I tested all four situations two times, and there was no difference between the two results.

Today I wanted to mesure everything once more, just to be scientific. It did not work out as expected. Today all values were about 9-10W higher for both machines idle and loaded, which means strong systematic error in the power meter.

Conclusion: this power meter was precise enough to be able to state, that a P4 consumes more power as a PPC G4. But nothing more, and we already knew it...

Does Pegasos/Linux support Nap mode for G4?

Hi Gunnar and Neko,

today I tested the power consumption of the 1st generation ODW with 1GHz 7447 G4, 40GB Seagate 7200rpm HD, 256MB DDR266 RAM, Radeon 9200SE, a DVD-ROM/CD-RW drive (exchanged the Samsung for an LG equivalent) and the standard 250W power supply.

Standby Power: 3-4W
Booting into MOS, idle: 48W
Quake3 timedemo four: 54W.

So, if you need a single number, I would say 50W sounds about right:
Dnetc rc5 (without AltiVec support in MOS) did not put much additional load on the system (less than the 3D demo), just like playing videos or copying data from DVD+R to HD.

After terminating the Q3 test the idle wattage sunk to an indicated 44W - maybe the difference to the initially shown 48W is due to a deviation in the power measurement device ELV EM94 (ad hoc test without calibration).

Hi tarbos, thanks.

That is good info.

Sounds like it is time for the second PegasosPPC Day. This time with the OSW. How many people do you think would like to be there?

R&B

_________________
http://bbrv.blogspot.com

Sign me up! Hehe..

It shouldn't be too hard to get 50 people to attend.

Will OSW feature a Hypervisor for virtualization and ressource management?