Some Insight Into the HMC-to-SDMC Transition

May 2, 2011 Timothy Prickett Morgan

Two weeks ago, I told you about how IBM was replacing the Hardware Management Console (HMC) for rack and tower Power-based servers with a new console called the Systems Director Management Console (SDMC) that would add BladeCenter chassis and Power-based blades to the mix with converged Systems Director-HMC functionality.

I received an interesting comment on the SDMC from a reader of The Four Hundred and also stumbled on a roadmap for the HMC-to-SDMC transition that spreads a little more light on Big Blue’s plans that its announcement April 12 did.

First, the reader comment:

Is it just me or is this a bit steep as a requirement for a management client? “If you run the SDMC as a virtual appliance on top of ESXi or KVM, you need a quad-core processor E5630 processor running at 2.53 GHz (or a faster one), 6 GB of main memory, and 500 GB of disk capacity.”

I think if it was my VMware farm, I’d be thin-provisioning that since I know I can run the HMC 7.7.2 in VirtualBox on my laptop and it’s as fast as the 7310-CR3 or CR4 with only 1 GB RAM and a single core assigned and it only needs around 10 GB for a base install of the code. Maybe it slows down a lot when fully loaded though, otherwise, if I was a cynic, I’d say that those specs are solely to make IBM’s sale of an entire server to run the software seem reasonable. Luckily, I believe everything IBM tells me, so I’d never suggest that:|

–Richard

The hypervisor console racket is just that, a racket, and all of the suppliers are milking it. In the case of the hypervisors from VMware, Citrix Systems, and Red Hat–that’s the ESXi, XenServer, and KVM hypervisors, respectively–all of the functionality is actually built into the hypervisor and the console is merely used to allow access to it and to do configuration and monitoring. Over at VMware, you shell out $4,369 per socket for vSphere Enterprise Plus, the basic usable, enterprise-class version of ESX with live migration, thin provisioning, and other features you need. The vCenter console costs $6,244 just for the software, and if you want to cluster together a bunch of vCenters to make a private cloud with a self-service portal, capacity planning, and the vCloud Director fabric, add another $325 per virtual machine on top of that. When you do the math, it takes about a third of the budget to buy fat servers for a private cloud, another third for the basic hypervisors from VMware, and another third for the vCloud Director extensions. And we’re talking about around $1,000 per VM. And that doesn’t include the cost of operating systems, middleware, databases, storage, and networking. That is virtualized servers with cloud extensions.

It’s good money, if you can get it.

Now, back to the SDMC. Here’s what the roadmap looks like:

As you can see, IBM eventually plans to move the SDMC microcode to a Power-based console in the second half of 2012, which is an interesting development, making it truly an all-Power solution, top to bottom. Support for the future Power7+ servers is also expected in the SDMC in 2012, if the scale on this roadmap is correct. (And you have to be careful assuming that it is.) Around the same time, the HMC microcode will be tweaked to support the Power7+ machines as well as Single Root I/O Virtualization, a PCI-Express bus standard that allows multiple operating systems to share a single PCI peripheral device. Obviously, the HMC or SDMC needs to control how this works.

The Integrated Virtualization Manager (IVM), a baby console that IBM created to spawn simple Linux partitions on Power-based machines, is embedded in the Virtual I/O Server (VIOS), which is itself a cut-down AIX kernel made to run peripheral drivers n that can be accessed by other logical partitions on a Power-based server. IVM works on Power7 servers and will be enhanced to support Power7+ machines, but its future beyond that seems to be in question. IBM has not yet decided if IVM will work on future Power8 machines. The HMC will not work on Power8 machines.