IBM’s Power-Based SmartClouds on the Horizon
April 11, 2011 Timothy Prickett Morgan
If you have been waiting, like I have, for IBM to get Power-based infrastructure clouds to market, it looks like you are going to have to wait a little while longer but no later than the end of this year. The Global Technology Services division of Big Blue’s Global Services giant, which accounts for more than half of the IT giants sales these days, fluffed up the Smart Business Cloud Enterprise last week, IBM’s first true infrastructure cloud.
The Smart Business Cloud – Enterprise, as IBM’s first true infrastructure as a service (IaaS) cloud is called, is based on the X64-based CloudBurst pre-configured stacks of servers, switches, and storage that Big Blue first announced back in June 2009 and that IBM upgraded with shiny new X64 machines and Power7-based stacks last October.
The SmartCloud Enterprise edition (IBM is dropping the “Business” and the spaces in the product name when it is talking fast) that was launched last Thursday is available now, but only for X64-based servers. Tim Koundais, director of cloud computing at IBM, says that SmartCloud Enterprise will come with a service-level agreement (SLA) that guarantees at least 99.5 percent availability, which if you do the math works out to something akin to 43.8 hours over the course of a year. That’s better than the SLA guarantee that the basic Amazon EC2 compute cloud has, which is zilch. If you want an SLA or tech support, you need to pay Amazon Web Services, the subsidiary of the online retailer that runs the EC2 cloud, extra for it. IBM assumes that the SmartCloud will be used for business, and hence an SLA is required. And Big Blue likes to have these things in writing.
IBM is using VMware’s ESXi 4.1 hypervisor as the server virtualization abstraction layer for the SmartCloud Enterprise IaaS cloud; the Rational-branded development and test cloud that Big Blue announced last March used Red Hat’s then-new Enterprise Virtualization hypervisor, which is based on the open source KVM project that Red Hat gained control of when it bought Qumranet a few years back. VMware is not cheap, but it is clearly the hypervisor of choice inside corporate data centers, so IBM has to start here so companies can take the server images they have created internally and deploy them on the clouds. It makes sense that as customers want Xen, KVM, or Hyper-V hypervisors, IBM will be able to partition off chunks of the SmartCloud infrastructure (and do so on the fly) to run whatever hypervisors customers require. As long as enough people are willing to pay, IBM will be willing to charge. And if the numbers are small, or the grief factor is high, then IBM will pass on the costs to users.
Because of the precedent of completely transparent pricing set by Amazon with EC2, IBM’s Global Technology Services has to provide a list price for the SmartCloud Enterprise cloud, and it did so here. IBM has five different configurations of 32-bit and 64-bit virtual machine slices, which are named copper, bronze, silver, gold, and platinum. You can see all the configurations here.
IBM is not saying what the underlying processors are in the hardware, but is carving up the machinery into virtual 1.25 GHz machines. With 32-bit versions of Windows or Linux, a VM can have one, two, or four of these virtual cores, from 2 GB up to 4 GB of virtual memory (the maximum that a 32-bit machine can address), and from 60 GB to 350 GB of dedicated storage. On 64-bit virtual machines, IBM’s server slices range from two to 16 virtual cores, 4 GB to 16 GB of virtual memory, and from 60 GB to 2 TB of disk capacity. You can add extra persistent storage to each virtual machine beyond this in blocks of 256 GB, 512 GB, and 2 TB. You can reserve static IP addresses for the server images, and IBM can also set up a virtual private network.
On the Copper configuration for 64-bit machines, a Windows Server 2003 or 2008 slice costs 34 cents per hour, SUSE Linux Enterprise Server 11 costs 35 cents per hour, and Red Hat Enterprise Linux 5.4 and 5.5 costs 40 cents per hour. As you move up through the virtual server sizes, Windows slices get more expensive faster than RHEL images, and SLES images are always cheaper. (I have a feeling this reflects IBM’s actual costs for operating system licenses.) On 32-bit images, Windows and SLES 11 are significantly less expensive on the Copper and Bronze configurations. If you want to reserve capacity on a monthly basis, IBM charges an up-front fee of $1,300 for 12 months and $1,850 for six months and then gives a slight discount off the per-hours charges. There are charges for data transfers, reserving IP addresses, and VPN services, and IBM also charges a premium if you want Big Blue to manage the operating system image, keeping it all patched and up to date. That OS management costs 10 cents per hour for RHEL and Windows and 8 cents per hour for SLES.
The SmartCloud Enterprise offering also has a mix of IBM’s WebSphere middleware and content management software, Lotus Domino groupware and application server, and DB2 and Informix database images that can be loaded from the SmartCloud self-service catalog and plunked onto images. The price you pay to use this software on your images depends on the size of the slice you buy, with a low of 12.2 cents per hour for Lotus Domino on Linux on a Copper image to high of $32.1 per hour (that’s dollars, not cents) for the WebSphere Portal Server and Lotus Web Content Management combo. There are development releases of DB2 Express-C and Informix Developer Edition that are free on all images. It is not clear what the real DB2 and Informix databases will cost, since they are not on the price list.
In the second half of this year, according to Koundais, IBM will roll out the SmartCloud Enterprise+ infrastructure cloud, which will give customers a choice of either X64 or Power7 iron. With the Enterprise+ edition of IBM’s IaaS cloud, IBM is guaranteeing 99.9 percent uptime, which works out to about 8.8 hours of downtime per year. SmartCloud Enterprise+ will be a multitenant cloud that has several different levels of security isolation, according to Koundais. IBM didn’t say what Power7 iron would be used in this future cloud, but it would be totally silly if it was not the Power 750-based CloudBurst machinery IBM announced last October. The Enterprise+ IaaS cloud will be fully managed by IBM Global Services and will be sold under fixed contracts or on monthly usage charges. You won’t be able to get this on an hourly usage schedule like the Enterprise edition cloud.
Presumably there will be an Extreme Enterprise Edition that has higher availability than even the 99.9 percent uptime guarantee of the Enterprise SLA, perhaps including actual live migration of VMs or logical partitions between physical machines. It is hard to imagine that IBM would offer an IaaS cloud with less than 99.5 percent uptime, but it could.
The SmartCloud Enterprise+ cloud will use VMware’s ESXi as the virtualization layer on X64 iron, which will support the same Windows and Linux options as the Enterprise Cloud. The Power-based servers offered under the Enterprise+ cloud will obviously use IBM’s PowerVM hypervisor to carve up logical partitions on the machinery. IBM says it will be supporting AIX and Linux images on the Power side of the Enterprise+ cloud, and that it understands that support for the IBM i operating system is a “requirement” and is “on the roadmap.” I read IBM the riot act about showing the OS/400 and i base some love, as you would expect, and hopefully it won’t take too long to get some sort of live migration feature into IBM i to make it truly cloudy.