Workload Partitions Not Coming to i5/OS V6R1?
July 30, 2007 Timothy Prickett Morgan
Last week, in the wake of a server reorganization that saw the System i business split into two bits, IBM announced an upcoming Power6-based, enterprise-class System i 570 server and also previewed some of the features in the upcoming i5/OS V6R1 operating system, due in 2008. IBM walked the midrange press through a rough sketch of the July 24 announcements to get the word out to the i5/OS and OS/400 base. But two features–workload partitions and its related application mobility–were not on the list.
IBM first started talking about workload partitions and application mobility back in early 2006, when VMware was getting a lot of good press for its ESX Server virtual machine hypervisor for X86 and X64 servers. ESX Server is a hypervisor similar in concept to the Virtualization Engine hypervisor that IBM created for its Power-based servers; both allow multiple and sometimes incompatible operating systems to run side by side on a single physical box within a virtualized hardware environment. One of the neat things about ESX Server is a feature called VMotion, which allows a workload running inside a VM partition on one machine to be teleported while the whole workload is running to another VM partition on a second physical machine. The two servers have to be linked to the same storage area network to accomplish this task, and once you know this, you can understand how VMotion works. Most of the data relating to the VM partition that is to be moved is on the disk drives, with only a relatively small portion describing the state of running programs in main memory. VMotion captures the state of main memory, packages it up, ships it across the network, changes the pointers to the SAN for that VM partition to another VM partition on the second machine, grabs the memory state, and loads it in that new partition. Voila! The workload is running on a second machine. The trick is to move that memory data in real time in such a way that end users do not see a blip in application performance.
IBM was expected to provide functionality similar to VMotion with a patch to AIX 5.3 in the summer of 2006, but that didn’t happen. IBM said last summer that the patch with this new functionality was being put off until early 2007, and then early 2007 came around and there was still no Power6 servers and no updated AIX coming out the door. IBM has tweaked AIX 5.3 to support the Power6-based System p 570 server that started shipping in June, but this updated logical partitioning capability is not part of the update. And now, the workload partition and application mobility features are being previewed to customers in a beta of AIX 6.1, which is slated for delivery this fall–maybe in October or November, if the word on the street is right.
As we reported back in May in our sister publication, The Unix Guardian, workload partitions, or WPARs, are a little different from the logical partitions, or LPARs, currently created by IBM’s Virtualization Engine hypervisor on Power-based machines. From the sketchy data I have on them, workload partitions seem to be analogous to a virtual private server on Linux and Windows servers as implemented by SWsoft‘s Virtuozzo hypervisor or on Solaris servers as implemented by Sun Microsystems with its operating system containers. With WPARs, IBM is providing a way to consolidate multiple AIX workloads into a single instance of AIX while providing security and administrative isolation between the WPARs. This is distinct from the current LPARs, which have a hypervisor layer riding between the hardware and multiple instances of an operating system–in the case of IBM’s System p and System i servers that’s AIX, i5/OS, and Linux riding above the hypervisor. While IBM can have a single operating system instance span from a fraction of a processor core to all of the cores in a box using LPARs, each LPAR has its own full instance of the operating system, which can be a headache when it comes to upgrading and patching virtual servers.
With AIX 6.1, IBM is creating a tool called the Workload Partitions Manager, and it will be used to provide live migration of AIX partitions between two physical Power-based servers as described above. IBM actually demonstrated an Oracle 10g database migration on two Power6-based System p 570s linked together using a 10 Gigabit Ethernet network and running a beta of AIX 6.1 at the May launch in London for the new machine and the preview of the Unix operating system.
Which brings me to i5/OS V6R1. I had been hoping that such functionality as WPARs and application mobility would be coming in a future i5/OS release–namely, i5/OS V6R1, now that we know its real name. But Craig Johnson, a product manager in the newly created Power Systems group and a long-time product manager from the System i division, said that IBM is not previewing this functionality. It was pretty clear that IBM doesn’t want to talk about this, and when I pressed for a little more detail, I was told that IBM has had subsystem-style partitioning inside OS/400 from the beginning and that this functionality was roughly analogous to the workload partitions that AIX is getting. This is certainly true in many respects. But there is one big difference:
As far as I know, you can’t move the jobs running in subsystem or even the whole collection of jobs running in a subsystem from one physical machine to another.
I think that this functionality would be very useful to a lot of customers. Of course, workload partitioning might take away one reason to install high availability clustering software on the i5/OS and OS/400 platform, which is something that none of the HA players in the IBM midrange want to have happen. But, then again, a lot of the guts for HA clustering are in OS/400 and i5/OS and those HA software providers really offer innovative ways to use and manage it, so you could make the same argument for workload partitions and application migration in i5/OS.
While subsystems are physically locked down to their machines–or as much as is possible in the highly virtualized AS/400-iSeries-System i architecture, where if you get right down to it, nothing the operating system is seeing is real and that is the beauty of it–it should, in theory, be possible to provide application mobility with subsystems. Maybe this is what IBM is working on. Maybe not. No one is saying right now. But if there was an open beta of i5/OS V6R1, as there is for AIX 6.1, we would have a better sense of what IBM’s plans are and what is not in the cards.
Which brings me to the final point I wanted to make: Wouldn’t it be nice if IBM opened up the beta for i5/OS V6R1 to anyone with a Power4 or higher server? If the AIX folks can let go, then the i5/OS folks should be able to do it. The more people who test the code, the better it is, and the more ideas you get.