Data Centers

Where do you host your data and how is it stored? A mix of environmental engineering, location choice, security practices and oodles of storage and processing power with lots of bandwidth-what's new?

Sunday, December 18, 2005

Data centers have the following necessary capabilities: 24x7 availability, high-bandwidth, power and heat conditioning, good physical access security and sound location. These are meant to be locations that will and can store all of your information for as long a period of time as possible. Not all data-centers are purely storage, though. Data centers can host your processing power too. These can perform off-site computations like long range forecasts and so on, freeing up local resources for other everyday computational tasks.

Three events are upper-most in everyone's minds from recent history. The last Gulf War, the Asian Tsunami and the earthquakes. All three have forced people to re-think where they placed their data centers and how secure those locations were from future calamities-both man-made and natural. For instance, would you place your data center in a place prone to Tsunamis or inter-continental wars or a nuclear war? If you consider last year's Mumbai rains, would you place them where it's going to flood once in a blue moon? Actually, depending on how you defined your 'data center' and how significantly much it is different from someone else's 'server room', you may not have a choice at all in where you place it.

Consolidation
That does seem to be the word going around the industry: Consolidation. Right from equipment and software vendors to solution providers and consultants, everyone is busy implementing or advising consolidation of resources. But how exactly do you consolidate data centers? The answer lies in determining the goals of your data centers and consolidating their needs and making better use of the available resources.

Also, part of those worries can be outsourced or co-located with someone else. For instance, the least deployment would be a data center, a DR site.

Possibly, the DR site could be a couple of co-located servers and some storage with a specialist. This way, consolidation no longer only means how much equipment you have, but also includes short and long term costs. It has become in effect, a consolidation of resources that include your budget. The simplest example of consolidation is of course moving from tower to racks and racks to blades.

Talking of blade servers, the action now is on Itanium, Itanium 2, Opteron and Dual Core blades. This follows the release of the 64-bit version Windows Server 2003. Vendors like SGI, IBM, HP and Fujitsu have announced or released such blades.

Bandwidth
Has this actually been an issue for traditional data centers? Available bandwidth has increased and the associated costs have dropped a lot in the past year also. That's another incentive to centralize the data center.

Better design?
The past year has seen vendors going around shouting 'thermals'. When people switched from racks to blades, the need for cooling went up, because instead of just two servers in a 2U box, you now had 25 of them -50 of them if your blades were dual processor. And, these boxes need to be cooled side-on instead of vertically. Thus, the design of the racks and cabinets housing the blades has had to undergo a design change to accommodate the new cooling requirements.

Virtualization
Hardware and software virtualization plays a big part in resource utilization. And when you're talking about the costly irons in a data center, this is important. When a physical server is hardware partitioned, its resource utilization is better. While the past year has seen Intel talking about including hardware partitioning support in their new CPU chips, servers have had them for ages, on different platforms including the Power series from IBM. Add to this the reality of multi-core CPUs and 64-bit processing and you can imagine the raw power available on such servers.

Virtualization is also a key to consolidation, since it allows you to utilize what you have in so many different ways. For instance, if you had a single blade server (with 25 blades in it), you could add a virtualization layer to that server and make yourself a 25-CPU server.

What is key to the success of virtualization, is actually how the workload is managed. Unless workloads or resources can be dynamically shifted to suit conditions, the system will continue to face lopsided resource marshalling. For instance, the workload manager should shift more resources from a lesser-utilization partition to the one that's almost full up. HP's Adaptive Enterprise is one such approach. Similarly IBM has its Virtualization Engine that adds virtualization support to the i, p and x series servers. Sun has the N1 Grid.

However, all the action today in virtualization is happening with VMWare. Their updated line up of software packs more punch to the virtual machines they help create and host. For details, see the box on Virtualization. Competing software like Xen and Microsoft Virtual Server are trying to fight back, but in so far, their features leave them quite far behind. The issue of virtualization though is quite serious. Linux vendor Red Hat has already announced that its next release of Red Hat Linux (2006) will bundle native support for server virtualization! The specifics of what this would include are not available yet, although it is expected to give competition to full-fledged virtualization products from other vendors.

How will virtualization and consolidation affect your data centers? One emerging picture is that enterprises may no longer need costly full-fledged data center buildings, but can do with a single rack full of blades (that's about 250 blades in all) in a server room  somewhere in a corner.  That brings down long-term costs too, both in terms of required man-power to manage your servers as well as in costs of space (which is usually at a premium in high IT adoption density areas), power, etc.

Blade PCs
In the future, the enterprise desktop may be a dumb terminal, the software running in a blade somewhere in a data center. This might come true if IBM's Virtualized Hosted Client Infrastructure succeeds. The average ratio is about 15 virtual PCs to a blade. The ecosystem of software that does this consists of-a VMWare GSX Server provides the VM hosting, a Citrix Presentation Server provides the interface to access the VM. Pricing is not currently available although the first products are expected to ship early 2006.

Page(s)   1