More On That Dreamy And Flashy Power 720 P05 Machine
November 28, 2011 Timothy Prickett Morgan
Once again, many thanks to the intrepid and helpful reader of The Four Hundred who gave us some insight into the equipment purchasing alternatives facing anyone who buys Power 720-class servers from IBM and the forward-looking direction the company cutting the check for the new system took. (Notice the lack of his or her in that sentence? I had to bend it pretty far to do that.) To make the situation a little clearer, that helpful reader gave us some more insight.
To recap: The Dreamy and Flashy Power 720 P05 Machine is one with no disks and no expansion features that is more compact and has more I/O bandwidth thanks to flash disks than a fatter configuration using disk drives that requires external disk enclosures and expensive 240-volt power. It has four 3 GHz Power7 cores, 64 GB of main memory, 1 TB of solid state flash drive capacity, is rated at 23,800 CPWs, and better still, remains in the P05 software tier for the IBM i operating system and third-party applications. In this case, the machine also supported a peak 1,600 users using good old RPG and native record-level I/O to the database.
Here’s some more insight:
Hi again, Tim:
I see you did run the numbers on the configuration I wrote you about earlier. Did you ever try running the numbers for the disk configuration with the expansion chassis, GX++ card, disk controllers, and hard disks?
[Answer: No, I didn’t because I was nearly falling over of exhaustion and took you at your estimate in your original email to me, saying it cost $70,000 to $80,000 less than a Power 720 P10 configuration using disks. –TPM]
Looking back, I see we had looked at one feature 5630, which costs $1,800, eight feature 1888 at a cost of $798 each, one feature 5615 at $1,750; one feature 5796 at $5,440, one feature 7314 at $525; one feature 6446 at $76, three feature 5886s at $4,950 each; two feature 5908s at $8,500 each; two feature 1865s at $475 each; plus miscellaneous cables for about $500. For a grand total of about $65,000! These prices are all IBM list, so you know there will be some discount, but the other hardware configuration was discounted as well. Note that in this configuration, several of the expansion chassis require 240-volt power, and the first configuration [outlined in the original story] requires only 120-volt power. 🙂
The only problem I had with your write-up was at the end, when you started discussing cost per user. The 1,600 user number I gave you was the peak number, not the average per day. We probably average 600 users most days, hitting the highest number of users only a few days in a year; so most of the time we are running about a third the number of jobs we experience on our busiest days.
I think this one-third ratio is the crux of the problem trying to justify the cost of the system to management; especially if the extra capacity will push you into a higher software tier and incur the penalty on all users.
Because of this penalty, we actively seek ways to reduce the load on the Power Systems box, even though it could certainly be doing more work most of the time. The cost of the hardware and base operating system is really only a part of the total cost of delivering the application functionality to our users. To that we must add operational costs, development costs (amortized), and end user support and training. Because of the nature of our business, there is not a highly visible connection between “bigger, faster, more powerful” systems and the bottom line. This means we are viewed as a cost center, not a producing asset. Trying to get more money for server upgrades is difficult at the best of times, and you could imagine the reaction if we tried to justify adding a workload to the Power 720 that bumped us into the P10 software tier when we could have put it on a separate $3,000 Linux box.
We have to buy a system in the size increments IBM allows, whether or not we will fully use the next increment of capacity or not. We also have to purchase that increment even if it will only provide value one day a year. Until there is some pricing mechanism that will allow us to purchase the capacity when it is needed (and only as much as is needed), the customer will always be griping about how much the systems cost. We opted to spend essentially an extra $10,000 or so to get a performance cushion of four to five times our estimated current needs, knowing that predicting the future is a risky business. Had we gone with the standard disk options, we would not have been able to afford this cushion, as the P10 software component would have pushed that margin to closer to the $100,000 level, not $10,000. We would have ended up paying more (including unlimited 5250 users on only two P10 cores) for half the performance we actually received. Same box, same software, same functions, just slower and more expensive!!!!
Then you get to add the extra charges for P05 to P10 versions of the application packages and third-party tools to the price. . .
–Name withheld upon request
Thank you again for clarifying on the peak users versus normal users and all of this additional insight. I really appreciate it, and I know readers of The Four Hundred do, too.
If I redo the math on a per-user basis for this system based on normal concurrent users on that flashy Power 720 P05 machine, at $114,441 per box at list price, which works out to $190.74 per user. That’s still only a little more than 52 cents per day for a year, and only a little more than 13 cents per day if you amortize it over four years–and you have plenty of capacity to handle that peak 1,600 users when those big days come around.
I just love that this is a real-world example, not some theoretical thing I cooked up in a spreadsheet. And I also love that you can make it sit up and bark using RPG, DDS, and record-level I/O.
Just for a laugh I checked to see if this flash configuration will go in the 2U Power 710 P05 machine as well and it will as long as you don’t mind using the HEA for Ethernet connectivity and adding a feature 7216 “media tray” with a SAS LTO tape in it.
So that’s a 3U solution if rack space is really tight 🙂
P.S. Thank your eagle-eyed system buyer for highlighting the feature 2053 SAS SSD controller. It’s pricey but, as you point out, still competitive with the P10 system once you include expansion and external disk trays.