TCA/TCO Study on Lotus Domino x86 consolidation to IBM Power Systems

July 26 2010

This popped up on a Google alert for me over the weekend... I am honestly not sure why it is buried in a channel announcement deep in the IBM website, and why the PDF seems to crash my browser.  Will see if I can find the right person to discuss both.  Assuming you get to it and open it in Acrobat, it's a very cool paper...

In the following study, we'll explore details for example of savings of almost a $1M Total Cost of Ownership (TCO) by running Domino mail on one Power 750 with 12 cores running at 57% CPU utilization instead of using 41 HP ProLiant DL380 G5 systems with a total of 164 cores running at 17% CPU utilization.

Consolidation savings include 85% in software maintenance, 95% in floor space and 96% in energy costs. We will also see more than $500,000 in TCO savings using Power when compared to consolidation using 20 HP Nehalem-EP based systems.
Link: TCA/TCO Study on Lotus Domino x86 consolidation to IBM Power Systems >

Ed Brill | Highland Park, IL USA Category:  notesdomino lotusnotes 8.5 lotus  Add/View Comments [5]

Post a Comment

  1. 1  Thomas  | 7/26/2010 11:55:47 AM

    Hi Ed

    I was excited to read this TCO study, but was disappointed about the result. Like any other "study" its just a theoretical exploration, and not a real life scenario.

    To host 30k Users we do not need more than 12 IBM x3650 M3 machines, which incl. storage are very well below 500k$.

    I'm sure there is an advantage of using Power7 technology, but the way it is presented here is not really what I had expected.

    Authors are talking about planning for growth. Okay, a powerful system can support that, but how about a company which is reducing the number of employees, I assume IBM cant cut one machine into two pieces to reduce TCO. with smaller machines you can always switch one off, with a big machine you cant.

    I know you'll say "you can use the capacity for something else", sure... but lets be realistic.

    ...furthermore, can anyone tell me why its always an HP hardware in the alternatives? Shouldn't you propose to use IBM hardware???

  1. 2  Mike Robinson http://www.invcs.com | 7/26/2010 12:54:25 PM

    Yeah I would have say 41 x86 boxes for 30K users is a bit over kill. I do like the concept though and I wish these things would play out in the real world as they stated in the studies. The one thing these studies never account for is staff retraining. It seems to be hard for an Admin w/ Windows based skills sets to become as proficient in something else (linux, aix, whatever Power7 runs, etc.).

    I recently saw a Domino Admin w/ Windows skills who now also has to manage AIX lpars w/ Domino dpars freeze in intimidation when trying to configure Symantec AV on said AIX boxes.

  1. 3  Rich Hunter  | 7/26/2010 2:35:36 PM

    Actually 41 servers for 30,000 users not over kill at all. That is an average of 731 users per server and does not account for clustering and fail-over. We assume that all out users can and will login at the same time on any given server and expect they have reliable response time when they do.

    The fact that one AIX box can handle the load is simply amazing and shows how well domino code can thread and scale. We run Domino on Solaris and see the same type of performance gain over x86 systems. Also the 64bit Unix systems seem to be quite a bit more mature than the x86 64 bit systems. Running Domino 32 bit on 64 bit Solaris has worked well for us for over a decade.

    Finally when you take into account the average size of the mail databases and large attachments and the index process, the projected numbers seem correct.

    When you take into account the monthly security patches for Windows and the time it takes to do that on 41 servers versus 1, the TCO looks good for this solution as well.

  1. 4  Tripp Black http://www.mindwatering.com | 7/26/2010 9:56:56 PM

    @3

    For 1/2 of my customers this is exactly right - assuming most are active sessions ...

    1000 concurrent users on a dual core server or VM with 3GB is about right. (2GB if on Linux w/o a graphical interface (GUI)).

    However, I learned something a few years ago - you don't wanted "mirrored" clusters. I personally like the 2 server "mirror" clustering because it's simple. But as I was shown by a few of my larger clients over the years, clustering in 3-5 makes more sense because you have users on 2 of the 5 servers and it begins to act more like RAID 5 than RAID 1 since the users have one primary server and their secondaries are "scattered" on the others. Typically, they use policies and OU units to "scatter" the additional/failover mail server. The failover load isn't 50% anymore but something like 10-20% which means you don't have servers with 1/2 load but a 60-70% load. Add VMs with CPU reservations and you're basically doing the same thing for less. This a the way to meet/beat the Power example above with our duel and quad core IBM and Dell eServers.

    If they are not concurrent users, then obviously it doesn't matter. You can put as many as the NIF (Notes Indexing Facility) can keep up with on the box. (That's anywhere from 5000-12000 on a box since only still about 750 - 1000 are active sessions.)

    With either model, you have about 800-1500 users on a server that are active and another 5000 or more on each box as a back-up mail server. The limit is typically the NIF.