« Today's Potpourri | Mainframe Home | Mainframe on YouTube - the sequels are in! »
Big Blue Going Green
When you click on a link, a server in a datacenter somewhere gets the job of finding the web page or process you requested and delivering it to your browser over the Internet. One user on the Internet and one server at the other end serving one web page is quite trivial. With millions of users around the world visiting the web site at unpredictable times and making unpredictable requests for millions of documents, pictures, music, videos, processes and transactions, it can become a nightmare for the people who are managing the datacenter. In the last five years there has been a six-fold increase in computing capacity and a 160 fold increase in storage. Along with the increase in capacity comes a huge increase in complexity and in electrical power usage.
Imagine looking through a window into a corporate datacenter (even though many of them are underground and have no windows) and you would see thousands of steel boxes mounted in six-foot-high racks with cables everywhere. This part of the problem has been addressed by new technology called virtualization, pioneered by IBM decades ago but greatly refined in recent years. (See "Virtually Real or Really Virtual"). Imagine a virtual datacenter. When you peer through the window you see three boxes -- a server, a disk storage device, and a network card. There is a person at a large video console who is looking at what appears to be a dashboard. It shows a pictorial diagram of all the things going on in the datacenter. When one application area needs more server, storage, or network capacity the virtual datacenter automatically re-allocates capacity from another application area that currently has excess capacity. The virtual datacenter keeps resources balanced, and when a component fails, the virtual datacenter automatically allocates a spare or underutilized component to take over. Virtual environments allow a big reduction in complexity but the even bigger problem is the huge growth in electrical power. In many cases companies are not able to get the additional power they need either because the power company does not have the capacity or because the datacenter is not designed to accommodate the physical changes necessary. Even if the power was readily available there is a negative impact on the environment. Hence, Big Green.
IBM is redirecting $1 billion per year across its businesses, mobilizing the company’s resources to dramatically increase the level of energy efficiency in IT. The plan includes new products and services to enable IBM clients to sharply reduce data center energy consumption and make them more “green”. The problem is sizable. Big companies spend tons of money on power. In IBM's case it is a half billion dollars per year. The priority has been on getting the servers and storage that are needed to achieve various business results -- need another feature for the web site, throw in another server. Have growth in web visitors -- throw some more servers at it.
IBM is leading by example. One of their "green" projects is consolidating 3,900 servers onto 30 new top of the line mainframe servers. The result is not only more compute power but dramatically less use of electrical power and space. One of IBM's customers went from 300 servers to six. The University of Pittsburgh Medical Center consolidated 1,000 servers onto 300 and saved $20m in costs while freeing up datacenter space for more hospital beds.
Datacenters have been popping up everywhere -- most of them built before 2001. The datacenters are very large rooms full of many different kinds of equipment -- designed in the same way they were decades ago -- like a kitchen where the stove puts out more heat so you turn on the air conditioning to cool down the entire room. The chef is comfortable and others in the room are freezing. IBM is designing datacenters for customers where cooling "zones" are specific to the type of equipment in each zone. Green datacenters not only save space and energy but also benefits the environment overall. In the past the electric bill has been allocated as overhead to all parts of the company. Redesigns are saving many millions of dollars. With the huge growth of energy for the IT infrastructure the CFO is reallocating energy expenditures from general overhead to the CIO so they can see what IT is really costing.
IBM has made a sizeable consulting business out of helping customers understand their energy usage and then designing and supervising the building of new Datacenters and cooling equipment. Having overseen the construction of thirty million square feet of advanced space, IBM has learned a lot. The virtualization is helping a lot too. It can now optimize the use of servers around energy use. For example, as workload declines, perhaps at night, servers can be virtualized and "moved" to underutilized servers and then automatically turn off the servers that are not needed for a few hours.
(See other IBM Happenings)
| by John Patrick | October 7, 2007 in Innovation Permalink |
TrackBack
TrackBack URL for this entry:
http://www.typepad.com/t/trackback/454446/22240870
Listed below are links to weblogs that reference Big Blue Going Green:
Comments
I recently got to talk to Nationwide's top z/OS and Linux on z guys. They have virtualized 800 Linux servers onto two mainframes and are very excited about it.
Here is a link to one of the articles about their situation: http://www.informationweek.com/showArticle.jhtml?articleID=192202104
Posted by: Christian McArthur | Oct 8, 2007 9:23:56 AM
For the University of Pittsburgh Medical Center, was that a mainframe consolidataion or was it to a blade server? The numbers do not look as dramatic and the link is broken.
If it is mainframe, as a Pittsburgh resident, I am glad to here something close to home. The other day a Sales person looked at me like I was crazy when I said that this was going to be a new trend.
Posted by: Steve Styer | Oct 10, 2007 2:07:04 PM
First, I am by no means an expert on z/VM and zLinux. However, in my reading, not everyone will have the results of Nationwide and IBM (and others). The number of Linux servers you can virtualize will be dependent upon the number of processors (particularly IFLs) on your mainframe (this is obvious) and also the workload of the Linux servers.
I can likely take a bunch of web servers and printer servers and put them on a mainframe than a database server or application server. The advantage of z/VM is taking multiple physical servers that are being underutilized and getting them to share the same hardware (even if they are running different configurations).
So (long story short) perhaps the Pittsburgh Medical Center's Linux servers were already doing a lot of work and was harder to share the same mainframe.
Posted by: Christian McArthur | Oct 11, 2007 7:33:44 PM
I found this article on the IBM case study http://www-306.ibm.com/software/success/cssdb.nsf/CS/JSTS-6K9UAT?OpenDocument&Site=gicss67hlth
It does mention a z-server in the mix.
Posted by: Steve Styer | Oct 12, 2007 7:31:06 AM
I wonder if anyone has seriously considered that stupid deployment of workloads and lack of systematic shared workload environment understanding is the origin and essence of these cancerous tumor farms of servers. Maybe smart architecture and real business understanding would yield more cost effective higher utilization of commodity linux servers rather than covering up that stupidity by jamming poorly architected workloads onto a mainframe.
Oh, and before your knee jerks and you assume I am an anti-mainframe bigot, please be assured that I think zVM, TPF, and zOS are incredible operating environments which have lots of lessons to teach the unixheads. They've had a few years to learn and still remain clueless, and now we get to see linux turn into a great way to burn up mainframe engine mips through classical unix-style inefficient use of mainframe architecture. IBM can take these wasteful engine usages to the bank- what a windfall.
It would be nice to see some critical and reflective comparisons that involve substantial rewrites of apache, mysql, and other core favorites to take advantage of mainframe strengths versus leaving them as is to run inefficiently. Doubt this will happen- gotta sell those engines you know.
Posted by: emes | Oct 17, 2007 12:07:49 AM
Hello! World!
Posted by: Hello! World! | Oct 22, 2007 10:42:14 AM
I always thought that IBM going green is a bit like a 4x4 saying its eco friendly lol.
However lower power consumtion will be welcomed in the UK where energy costs are far too high
Posted by: Webb | Nov 8, 2007 3:21:27 PM
Upfront - I work at IBM, selling support for HW & SW. I see customers every week that have server sprawl, for various reasons - decentralized buying, political & organizational structure, domestic vs international, marketing vs manufacturing vs engineering etc.
I have yet to work with a company that has a long-running overall strategy including standards for HW & SW acquisition, provisioning, deploying, sharing and supporting (including budgets) designed to operate optimally. If standards like that exist, they seem to have been created as the result of a trigger event like a merger, acquisition, financial trouble or outsourcing/insourcing. I think it typically is an evolution based on it being easier to run "one's own" hw, sw, and/or applications within departments or otherwise within a group of "silo" budgets. More of a "fail to plan" situation, not a "plan to fail".
Customers of mine who do tackle this usually start with HW, either through consolidation, virtualization or both. They are achieving significant savings in floor space, electrical, cooling, capital investment in HW, SW license fees, and ongoing support, including maintenance. Results vary widely as mentioned above, but even a 15%-20% savings across all the above items certainly adds up. And running a more centralized organization with centralized support can provide better data with which to manage the business - yielding more future benefits. Whether they rewrite application, middleware, or other code would certainly improve efficiency and effectiveness, but sometime you just have to start somewhere.
And I am not a hw platform bigot or even an IBM hw bigot, I guess maybe I am a "Services" bigot. What HW or who made it doesn't matter to me. Good Luck.
Posted by: Bruce | Nov 20, 2007 6:08:54 PM
As the de facto administrator of the Security Fix blog, I've spent many an hour deleting spammy links left in the comments section -
- comments that usually lead back to the same kinds of Web sites you most commonly see advertised in junk e-mail.
Posted by: Garri Azz | Feb 16, 2008 2:08:07 AM
lged asydc mbqdi nsqrjovc eayzbgnoq dgecs nmvfrxjz
Posted by: whbk jseicypn | May 13, 2008 6:35:03 AM
