Tuesday, September 30, 2008

Power Supply Energy Efficiency - Active PFC vs no PFC

Well, I've heard the chatter that the new PFC Power Supplies can save power, and I thought I'd give it a go.

I took a digital watt meter, and booted up a bench system with a standard ThermalTake TR2 430W non-PFC power supply. It was consuming ~120 watts under load (boot or stress test) and ~98 watts during idle.

I switched to a Seasonic S12-II 330W power supply with Active-PFC, and ran the same tests again.

With the new power supply, it was burning ~84 watts under load, and ~50 watts on idle.

I tried the same test on a few different motherboards, with some Enermax power supplies (also non PFC) and I've seen a significant drop in all. Another system that was 150w under load dropped to 109W with the Seasonic S12-II power supply.

A few items to consider:

1) There may be a slight drop in energy consumption because I'm using a 330W power supply vs a 430W power supply - but I expect it to be very slight, maybe a few Watts.

2) It could just be brand differences - I don't have a Seasonic Passive PFC or non-PFC power supply to compare. Running this test with more controls would make for a more accurate reading.

3) I've heard Passive PFC is more energy efficient than Active PFC, but I don't have the power supplies to compare.

4) The more power you burn, the more of a difference an Active PFC equiped power suuply makes. This makes good sense, as efficiency is a % rating of total power consumed - As you consume more power, your % lost to heat is greater.

5) I tested a cheaper Seasonic 330W power supply (generic OEM replacement, not the S12) and it had nearly identical stats for energy consumption as the S12-II, so you don't need to buy the more expenisve power supply to reap the savings.


I'm starting to phase out my non-PFC power supplies in my lab, because this does add up to a savings. I have ~20 machines in the lab, and about 10 of them are on 24/7.

I'm able to get power supplies at cost, $35 for a generic Seasonic, and $45 for the S12-II 330W, so this doesn't cost me very much. I'm also able to sell off my old power supplies as used for nearly the new cost of the Seasonics, so this will save me money if we don't count the time I have spent.

Otherwise, you'll have to take a serious look at if it's going to make $$ sense to switch your PS. In the best case scenario from my tests (a converted system running wide-open) is saving ~36 Watts. We pay ~.11 cents per killowatt hour here, so I'm saving ~.10 cents per day, or ~$2.85 per month.

That means it's over 15 months to pay back a $45 cash investment in this power supply, so your ROI isn't amazing. If you're not getting any money for your old supply, you are also throwing away the $$ you spent for that, unless it was time for a new PS.

On the other hand, when you consider that you'll probably have this solid Seasonic power supply for 5 years, and electricity will only go up in price, then it may make good sense.

Lastly, don't forget to account for the savings in A/C if you use it - You're making less heat now, which means less electricity to cool your office or home. In my lab, I think this will make a big difference, as on hot days we can just barely keep the temp from rising. When I unlock the lab doors in the morning before the A/C has turned on, I can always feel the heat from the room from these running systems.

With a lower heat generation load in the building, we should be able to keep our existing A/C system, which is a large cost savings.

Saturday, September 27, 2008

Speed Increase - ESXi Update2

Has anyone else noticed a speed increase since updating to ESXi Update 2?

I have.

I also notice a further speed increase with the 1006 version of BIOS for the ASUS DSEB board I use.

I did make some speed tests to show the increase, but I'm too lazy to post this morning - Just throwing this out there to see if anyone else has noticed it.

Thursday, July 17, 2008

RAID Alignment and VMware ESX 3i 3.5

Make sure to create your datastores from within the Virtual Infrastructure Client GUI, don't take what the ESX install CD creates (this means deleting any data stores created by the install CD).

It's because the Install CD creates the partitions on the standard Sector 63 starting position (31.5k into the drive. This isn't a proper multiple of any common RAID stripe size (2k, 4k, 8k, 16k, 32k, 64k, etc) so the partition ends up being out of alignment with the RAID stripe elements, and READ/WRITE performance will suffer.

This is all documented in a number of VMware documents.

Creating the partitions in VCI makes them start on sector 128 (64k into the drive), which lines up properly.

Questions on what RAID stripe size and what Windows Allocation Unit size to use will be in another post.

Wednesday, July 9, 2008

Quick and Dirty - LSI 3041E (1064E) SAS PCIe Card vs Dell CERC SATA 1.5/6ch Card


I have a customer with a Dell Poweredge 1800. It comes with the Dell CERC SATA 1.5/6ch RAID controller. It was configured for RAID5 operation.

SAP was killing their server with a lot of SQL requests - They have grown quite a bit in the last year, and their simple little fileserver was now hosting exchange, SQL, and a lot of users.

All my testing pointed to the drives not being able to provide the CPU with enough data. Memory and CPU were always low, while the disk queue was through the ceiling all the time.

I've been doing a lot of SAS drive work lately, and I sold them on a basic RAID mirror with 2 146 gig SAS drives (Seagate Cheetah 15k, 32 meg cache).

I selected the LSI 3041E (1064E based) 4 port SAS card because it was only $190 or so, and I've had good luck with LSI products lately (and less with Adaptec).

After an install and format, I ran a very simple IOmeter tests (the default test set) for 5 minutes. This is what I got;


Operations/Sec 62.98 398.28
Megabytes/Sec .12 .78
Average Response Time(ms) 15.88 2.50

As you can see, the SAS drive array beats the old SATA drive hands down in every test. It's 6.4 times better - 6.4 times more operations per second (reads, writes), it can move 6.4 times more data in a second, and it takes 6.4 times less time to respond to an operation.

Of course, this is a nasty test - It's comparing 7200 RPM 1.5 Gb/sec SATA drives to SAS 15000 RM 3G/sec, it's comparing RAID5 to RAID Mirror. Everything about it is wrong.

But it also shows how much faster life can be for $800 in parts i you have a similar setup. :-)

I'm waiting for real world user comments, and we'll do some SAP end user tests on a number of slow queries shortly to see if the this additional speed is actually being taken advantage of.

Sunday, February 24, 2008

VMWare ESX 3i 3.5 Speed vs Physical

(Note - Red is Physical Machine with 8 CPUs, Blue is ESX 3i 3.5 with 4 CPUs)

If you're serious about moving to Virtual Servers, you're going to want to know what type of performance hit you're taking for using this technology.

My rough, fairly unscientific results show that using ESX 3i 3.5;

CPU - About 5-10% less than physical

Memory - Nearly physical speed

Disk - Between 50%-75% of physical speed.

Disk takes the largest hit, so you best make sure to follow all the standard ESX "best practices" - Use SAS drives, put Swap on a seperate controller, etc, etc.

Details follow:

I have a new ASUS DSEB-D16/SAS motherboard that I have been running some tests on. So far the speed has ben quite good for ESX 3i 3.5.

My test platform:

- ASUS DSEB-D16/SAS Motherboard

- Two Xeon 2.0 GHz Quad Cores, 45nm

- 8 Gigs ECC Kingston Ram, 667Mhz

- 2 Segate 250 SATAII Drives, set up as Stripe

I've made some tests with Windows 2008 RC1 in both 64 and 32 bit modes, and with Windows 2003 64 bit mode.

I've always used the latest Window drivers for the board and controller.

I used PassMark 6.1 for my testing - It's not overly in depth, but it gives a good overview for what I'm wanting to do.

There is one major messup that I didn't expect - I can only run 4 virtual CPU's in 3i, not the 8 that I have. The first test was with 8 CPUs, and thusly the CPU speed is roughly 50% less in my 3i run. Passmark does make use of every CPU for the CPU portion of the tests (and it runs single CPU for the Memory and Disk mark tests), so this will affect the CPU portion of the tests only.

I have run the test again with only 2 virtual CPUs in 3i, and the results were roughly 50% less than with 4 CPUs. It stands to reason that wil a full 8 CPUs in 3i we'd be very close to the speed of the physical machine.

All my 3i settings are pretty default. I'm going to switch to SAS drives shortly, but I wanted to see the disk test speeds with the existing SATA 250s I used for Windows 2008.

So far, so good. I'm happy with the 3i speed. I can use 32 meg of RAM with my current setup, and it's less than $100 for a 2 Gig stick of ECC 667. I'll keep the RAM high so it's not swapping to disk, and minimize my performance hit on the drives with beefy SAS drives.

ASUS DSEB LSI 1068 MegaRaid Speed vs Intel Matrix

Okay, spent a few hours making some speed tests between the two built in controllers.

I have stats to back up my words, but I don't feel like dragging out all the tests I did and splicing up the graphics/results. I may get to it later.

In a nutshell, with Windows 2008 RC1 64 Bit on a ASUS DSEB-15/SAS Board with 2 Xeon Quads;

- Intel Matrix controller is ever-so-slightly faster for SATA drives than the LSI Megaraid, but not enough to really worry about.
- ESX 3i 3.5 does see the Intel matrix controller, but doesn't see the RAID sets created on it - Just the individual drives.

It looks like you'll still need the SAS option for this board if you want to run RAID of any kind in ESX 3i 3.5

Friday, February 8, 2008

VMWare ESX 3.5 and 3i on ASUS DSEB-D16/SAS Motherboard with SATA drives

Okay, the board arrived yesterday. So far both VMWare's ESX 3.5 and 3.5i are able to boot, install, and stay stable.

I'm not using the LSI SAS controller as of yet - Just the Intel Matrix controller, with a plain old SATA drive. I haven't created any RAID, so both versions of 3.5 are able to spot the Intel controller and work just fine.

This means that people wanting to save a few bucks could use the Intel 6321 controller only board (no SAS).

Watch out - Make sure you use SATA ports 1-4. Ports 5 and 6 are not spotted by the install scripts.


I've been making use of all the great technical information out on the net for over a decade now.

It's an invaluable resource to someone in the IT industry like myself, so I thought it was time to start contributing, to give back to the general net community.

I'll be posting various tidbits of information here that I think others may benifit from.