>Has anyone seen this NY Times article?  Perhaps the reporter should have 
>looked in the Times Tower for how they are saving power. Mainframes are far 
>more efficient, and these CIO's and companies are doing whatever they can do 
>to get rid of them. Maybe the Times needs a mirror....

Well, besides the "Mainframes are Better Argument" that Doug intended, there is 
something here to discuss that is interesting for both the mainframe and 
non-mainframe communities.

I can confirm that the data center power problem is extremely pervasive.  A 
relative who works for Intel tells me that this problem is a major focus for 
them.  The situation arose because of these factors:
(1) The very idea of shared infrastructure was foreign to those who built the 
eCommerce apps of the late 90's and early 2000's.  So if you were going to roll 
out a new web app you needed two or more new web servers, two or more new 
database servers, possibly two or more new application servers, etc.  And you 
also needed separate environments for DEV, TEST, and UAT.  So pretty soon you 
have filled up several nineteen inch racks.
(2) As CPU clock speeds kept rising, the TDP (Thermal Design Power) kept 
increasing too, hitting 150 watts with the Harpertown series of Xeon processors 
in 2007.
(3) The move to storing images, audio, and HD video have seen an explosion in 
demand for data storage.  Each disk needs 10 to 25 watts energy to spin as fast 
as 15K rpm (Seagate Cheetah).  But enterprise class drives have lower 
capacities that consumer drives.  So while we might have 1TB drives in our 
desktops, the enterprise RAID arrays are stuck at 73GB-146GB-300GB per HDA.

The end result is that even modest IT operations have hundreds of servers and 
disk arrays containing thousands of disks.  And most of these servers just sit 
doing mostly nothing.  And some servers exist simply because no-one can 
remember its purpose.

But change is coming, and not just with the Pie In the Sky idea of Cloud 
Computing.  Consider this:

(1) CPU TDP crested at 150 watts in 2007 and is now in decline.  The recent 
Sandy Bridge chips are all down to 90w or 65w.  Some are running as low as 35w.
(2) The industry has embraced Virtualization as part of the solution for server 
proliferation.  So instead of "Blade Servers", companies are buying big 
multi-socket, multi-core machines with tons of RAM and then using them to host 
dozens if not hundreds of virtual machines.  Some of these VM's will be very 
busy serving up web pages or accessing databases. Others will simply loaf along 
responding to occasional requests for LDAP requests, emails, etc. In other 
words, horizontal scaling is out of fashion, and vertical scaling is back in 
vogue.
(3) The CPU chip makers are about to apply Mobile technology to their Server 
CPU's.  Mobile CPU's dynamically adjust their clock speeds to fit the workload, 
sometimes running at 200MHz and at other times running full throttle at 2GHz+ 
or more.  Server CPU's will soon do the same.  When throttled down, a CPU uses 
much less energy and radiates much less heat.
(4) The spinning disk makers are about to also apply Mobile technology to their 
Enterprise disks.  No need to spin at 15K rpm when the requests aren't coming 
in.  They are also going smaller - down to 2.5 inches instead of the current 
standard 3.5.  Smaller platters mean faster seek times.
(5) NAND flash and NOR flash technology will soon start eating away at the Hard 
Disk market.  Already we have SSD's that are 1TB.  Supposedly we will soon see 
4TB and 16TB capacities, all with access times 10 times faster than HDD.  And 
with power consumption 10% or less.
(6) Server System Administrators are being overwhelmed with supporting the 
hundreds of server images, whether virtual or real.  So expect them to push 
back on the separate servers for separate apps practice.

While you might think that nothing here matters to mainframers.  But consider 
this:
(1) Server disk array technology is substantially the same as mainframe disk 
arrays.  Many companies use the same EMC arrays to provision both their Wintel 
servers and their z/OS platform.  If EMC arrays start using 2.5inch disks or 
SSD's, then the benefits will extend to both mainframe and non-mainframe 
environments.
(2) Many mainframes have gone to virtual tape technology, just as non-mainframe 
servers have done.
(3) Mainframers of course embraced virtualization long ago, beginning with 
VM/370 and later with LPAR.
(4) If you think that a z/Architecture processor is radically different from an 
Intel Xeon, you only need to read the respective descriptions.  So it isn't so 
far off the mark to think that z chips have the same issues as Xeon's and that 
the same fixes will be applied to both.

John

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to