IBM churns up CloudBurst clouds
Not raining on Google and Amazon parades
Cloud computing must be real - IBM is going to try to start making real money selling pre-configured clouds to enterprises for their internal use, targeting very specific workloads running on x64 servers to start and eventually encompassing its Power and mainframe systems.
IBM has been dabbling in and doing research on cloud computing for the past two years - including some joint work it has done with Google - and has years of On Demand utility computing under its belt before the more slippery and malleable cloud model stole the IT show.
You can debate about whether what IBM is announcing today with its CloudBurst cloud infrastructure and its related Smart Business cloud services are truly cloud computing: in a sense, they are not. But that said, what IBM is building and hoping to sell is one of the few ways that companies will probably feel comfortable moving toward cloud computing until they get a little more experience with this approach to running applications.
Cloud computing was supposed to not only encompass a specific way of acquiring computing and storage capacity - you pay for what you use, you get capacity turned on and off almost instantly to minimize the cost, and you have access to lots of capacity for when you need it - but also was, by some definitions, to be based on a more fluid and usually distributed kind of programming. Companies like to keep control of their servers and the applications and data that run on them, so even if they could upload their ERP systems onto the Amazon EC2 or the Google Apps clouds, they probably wouldn't. Maybe someday, but not today.
What Big Blue reckons IT shops will buy are preconfigured stacks of servers and storage that are integrated and easier to manage than piecemeal parts. IBM is not trying to sell aggregations of virtual server slices that are compatible with those being peddled by Amazon, Google, and others in their clouds, but rather giving companies the infrastructure and services to make their own internal clouds based on VMware's prior ESX Server 3.5 hypervisor and related tools.
It is only a matter of time before IBM supports VMware's vSphere 4.0 stack on its CloudBurst configurations; vSphere was launched on April 21 to much fanfare and started shipping on May 21. It will take months for server makers and end user companies certify vSphere, and given the substantial scalability and performance benefits of the ESX Server 4.0 hypervisor, you can be sure IBM wanted to launch its CloudBurst stack on this hypervisor. But so it goes in the server racket.
The basic CloudBurst hardware setup will look familiar to many of IBM's server customers. IBM takes 1 42U rack, chucks in a BladeCenter blade server chassis and a System x 3650 M2 server. The x3650 M2, which is based on Intel's new "Nehalem EP" quad-core processors, has 48 GB of main memory and is set up as a management server for the cloud infrastructure. There is also an HS22 Nehalem-based blade server in the chassis that is designated as a management blade inside the chassis, plus three HS22 blades to support ESX Server hypervisors and their operating system and application workloads.
All of the HS22 blades have 48 GB of main memory on them. There's room for another ten blades in that initial chassis for customers to expand their computing capacity, and multiple chasses can be added to the rack and multiple racks to the data center as the cloud grows. The blades also get a DS3400 midrange disk array that links to the blades through Fibre Channel.
On the software side, of this CloudBurst setup is VMware's ESXi 3.5 embedded hypervisor, which is stored on an internal flash drive that plugs into each HS22 blade. The System x management server and the HS22 management blade is loaded up with IBM's Tivoli Provisioning Manager V7.1 and Monitoring V6.2.1 as well as its Systems Director 6.1.1 system management tool, which includes active energy manager, which controls the power capping feature of IBM's servers.
The software stack also includes IBM's ToolCenter 1.0, DS Storage Manager V10.36, and an LSI SMI-S provider for the DS3400 array. The whole shebang is orchestrated with something called the CloudBurst V1.1 service management pack, the secret sauce that apparently turns this from a rack of virtualized servers and storage and into a cloud. Some of this secret sauce probably made it into the CloudBurst WebSphere deployment appliance that IBM announced back in May.
Next page: No more hanging around for tests?
It sounds familiar - again!
"Companies like to keep control of their servers and the applications and data that run on them, so even if they could upload their ERP systems onto the Amazon EC2 or the Google Apps clouds, they probably wouldn't. Maybe someday, but not today"
And yet many companies don't think twice about running their systems at off-site locations such as with IBM, EDS etc. - particularly mainframe stuff where it can be much cheaper.
As with most of these things, what is the norm in the mainframe world will trickle down to the smaller servers eventually.
WebSphere CloudBurst Appliance and IBM CloudBurst
IBM CloudBurst and WebSphere CloudBurst Appliance are two different offerings, with different price points. The price for WebSphere CloudBurst Appliance is $45,000 for the appliance and 1000 PVU of cloud capacity entitlement (the size of the pool of X86 machines being managed), and additional PVUs are $15 each. For example, a WS CloudBurst Appliance managing a 2000 PVU cloud would cost 45K + 15K = $60K.