ADSM-L

Re: [ADSM-L] Check signals on Power vs. x86...

2011-10-12 11:57:23
Subject: Re: [ADSM-L] Check signals on Power vs. x86...
From: Paul Zarnowski <psz1 AT CORNELL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Wed, 12 Oct 2011 11:50:21 -0400
We have kept with Power, largely because we (1) have AIX skills in-house, and 
(2) the effort to migrate to another platform would be great, because we have 
significant archive data.

Some thoughts (take them as you will):

1. You can scale power vertically, quite high.  We have 10 very sizable 
instances on two p750s.  Each p750 has many FC HBAs on it, which are shared by 
all instances on that server.  If we used intel instead, we'd likely have to 
purchase many more HBAs, or divide up our tape/disk resources so that they are 
not all sharable.  It is nice to be able to share everything with everything.  
We can keep adding processors and RAM for some time.  Higher-end p7's scale 
even higher, but we found the 750 to be a nice fit for our needs, giving us 
considerable head room.

2. I have two resource baskets to monitor/plan for instead of 10.  When I add 
RAM to a server, for example, it benefits all of the instances running on it, 
not just 1.

3. Power 7 has 4 SMTs per core, vs 2 for Power 6, and 1 for Power 5 and 
earlier.  If you look at the SPEC ratings, you will see that this translates to 
greater workload handling per core for Power 7 vs earlier Power, for the same 
GHz.  I am not sure what this looks like on Intel.  I would look at an 
appropriate SPEC benchmark when comparing platforms vs something simplistic 
like processor speed (GHz).  Speeds can be very misleading.

4. I am a fan of AIX's LVM and management suite.  I admit I may be biased, 
because I have lived with it for so long and know it, but as I have considered 
Linux, I have become aware of some things that it does not yet have, or doesn't 
have as nicely as AIX does.

I will be interested in what others have to say, as I share Allen's perspective 
on revisiting assumptions periodically.

..Paul


At 10:05 AM 10/12/2011, Stefan Folkerts wrote:
>I would like to see that as well, I find it impossible to believe without
>proof...and I love the power platform just not because it's 4x as fast (at
>least) per core as x86 for TSM because I don't think it is. :-)
>
>On Mon, Oct 10, 2011 at 10:36 AM, Gregor van den Boogaart <
>gregor.boogaart AT rz.uni-augsburg DOT de> wrote:
>
>> @Howard Coles:
>> > For every 1 proc or core
>> > of Power you would need 4 or more of x86 (even at their best level).  I
>> > have seen the numbers from Intel comparing Newer x86 processors to
>> > Power6 and they are just below the Power 6 (using 2x's the number of
>> > cores).
>> Do you have a reference, link, pdf, ... for this?
>>
>> > The problem is, You can get Power7 cheaper than Power6, and get
>> > twice the performance.
>> And for that?
>>
>> Thanks!
>>
>> Gregor van den Boogaart
>>


--
Paul Zarnowski                            Ph: 607-255-4757
Manager, Storage Services                 Fx: 607-255-8521
719 Rhodes Hall, Ithaca, NY 14853-3801    Em: psz1 AT cornell DOT edu