Original URL: http://www.theregister.co.uk/2009/11/02/ibm_ios_611/
IBM facelifts i/OS for midrange gear
Power system nip and tuck
It might have come out with a slightly different name and a little later than expected, but Big Blue has tweaked its proprietary midrange operating system for Power-based systems with the i 6.1.1 release.
The word on the street a little more than a year ago was that Big Blue would be packaging up some feature enhancements and patches and would roll out an interim i 6.1.5 interim release of the venerable OS/400 operating system (now known simply as IBM i) sometime in the first half of 2009, maybe in May, to be followed up with an i 6.2 full release in early 2010, perhaps in January. This information came out of people attending the i variant of the Power Systems Technical University in Chicago in September 2009, and the rumors were reasonably credible.
The rumors also turned out to be untrue, inasmuch as a rumor can ever be said to be true until a vendor actually commits to a release schedule and naming convention.
As part of the Dynamic Infrastructure marketing blitz this month, when IBM had no Power Systems server announcements to speak of and which was not quite blitzy enough to get people to stop talking about the insider trading scandal that has caught Bob Moffat, formerly the general manager of IBM's System and Technology Group, Big Blue released a little something called i 6.1.1.
The i 6.1.1 operating system began shipping to customers on October 23. As was expected more than a year ago, the point release of the operating system included enhancements relating to the Virtual I/O Server, which as the name suggests is a program that runs in a logical partition that is used to virtualize the I/O that other logical partitions in a machine (the ones doing real work) makes use of to talk to disks and tapes attached to the physical server.
Virtual I/O Server, or VIOS for short, comes out of the AIX side of the Power Systems house and is basically alien to OS/400 shops. But if you want to run i 6.1 on a Power-based blade, you have to use VIOS as the virtualization layer to talk to disk and tape and networks because i 6.1, for whatever reason, was not ported natively to the J series of Power6 and Power6+ blade servers. Ditto for selected external disk arrays, which can only be attached to i 6.1 through VIOS.
IBM uses VIOS so it can take some shortcuts. The company writes a driver for VIOS to talk to a specific Fibre Channel interface card, and then puts a virtual interface card in an AIX, Linux, or i partition that communicates with the real Fibre Channel card, but does so in an emulated way. The thing is, this is great for IBM, because it cuts down on the work it needs to do, but not so for i shops that have paid the price because they are forced to use a piece of software that they have no familiarity with.
Ditto for business partners. Which is why a year ago, IBM started preconfiguring VIOS and i 6.1 on BladeCenter setups with Power-based JS12 and JS22 blades. As I said at the time, it had not occurred to me that IBM was not already pre-installing this software on the blades. I mean, isn't the "i" supposed to mean integration?
Anyway, here were are in October, and VIOS, which is technically a part of the PowerVM server virtualization hypervisor toolset, has indeed got some enhancements. With the i 6.1.1 update, IBM is allowing machines based on the Power6 and Power6+ processors to support redundant VIOS partitions. Apparently, not only were i shops cranky about having another layer of software standing between their i-based logical partitions and their physical I/O adapters, but they were complaining that VIOS represents a single point of failure. So now companies can set up redundant active-active mirroring of VIOS using two logical partitions. You have to be at VIOS 2.1.2 to do the mirroring.
In another related item, the PowerVM hypervisor now supports an industry standard called N_Port ID Virtualization, or NPIV. With this support, i 6.1.1 can link out through native Fibre Channel adapters in a Power Systems rack or tower server (but not a blade) and talk directly to storage area networks. The upshot now is that IBM's high-end DS8000 disk arrays and TS3100, TS3200, and TS3500 arrays can be linked directly to i 6.1.1 and their full feature sets are available to the i platform; they are not being emulated in any way.
Invisible solid state storage
To use NPIV, customers have to link to a DS8000 from a rack or tower server using a Power6 or Power6+ processor, 8 Gb/sec Fibre Channel cards (features 5735, 8271, 8242, and 8240), and be running i 6.1.1. The tape libraries can link through the NPIV features from blades, racks, and towers, provided they have 8 Gb/sec Fibre Channel adapters linking out to disk and tape. Specifically, we're talking about the feature 5735 PCI-Express two-port 8 Gb/sec Fibre Channel adapter, the feature 8271 and 8242 QLogic 8 GB/sec Fibre Channel mezzanine card for BladeCenter blades, and the feature 8240 Emulex 8 Gb/sec Fibre Channel mezzanine card for the blades.
With the i 6.1.1 update, IBM is also offering the direct attachment of the midrange DS5100 and DS5300 disk arrays, which up until now have only been supported through VIOS on either rack or tower Power Systems or the J series of BladeCenter blades. The direct attachment requires a Smart IOA Fibre Adapter and will be available through a patch on December 4. IBM will be supporting the direct attachment of DS8700 disk arrays with i 6.1.1 with a patch that is due on November 12.
The DS8700s were already supported directly through Fibre Channel adapters on i5/OS V5R4, the prior release of the OS/400 operating system. This direct attachment for i 6.1.1 requires a feature 5749 Fibre Channel adapter (two port PCI-X with 4 GB/sec speed), or the more modern feature 5774 (a 4 Gb/sec card that plugs into a PCI-Express port) or feature 5735 (a PCI-Express card running at 8 Gb/sec).
On the solid state disk front, IBM has made some tweaks to make solid state storage a little more invisible to system administrators. With i 6.1.1, IBM is adding a feature called Disk Drainer, which moves data from one SSD to another inside of a Power Systems box. Another feature called Capacity Balancer does just what you would think it would do, which is balance the placement of application and data on one disk group in a system between the disk drives and SSDs in that group.
Most frequently used data ends up on the SSDs, which are orders of magnitude faster I/O than disk drives. IBM also says that the defragmenter used with i 6.1 - I didn't even know it had one, but of course it does - has been told to ignore the unscrambling of data on the SSDs and to just focus on the files on disk-based storage.
According to Craig Johnson, the i/OS product manager in the Power Systems division, IBM has tweaked the Storage Manager tool for i5/OS V5R4 and i 6.1 with new trace and balance commands to move hot data to the solid state disks in a Power Systems server. The tool has been tweaked and the DB2 for i database has been given an additional parameter (presumably as part of single-level storage) that allows DB2 objects to be placed on SSDs.
PTFs for i5/OS V5R4 and i 6.1 already allowed for SSDs inside of I/O drawers to be seen by the operating systems, and with i 6.1.1, the operating system can see SSDs out on SANs or under the control of VIOS partitions. This functionality will not be back ported, so if you need it, you need to move to i 6.1.1.
IBM is also providing i shops with a free SSD Analyzer Tool, which you can get here, that you run on your i5/OS V5R4 or i 6.1 system to simulate the effect that SSDs might have on the performance of your own applications.
Another neat new feature for i 6.1.1 is temporary file system support. Johnson says that lots of applications running on the i platform that make use of the Integrated File System have their performance degraded because the i/OS (like OS/400 before it) treats every object the same - as if you wanted to have it permanently stored. But for many applications, such as ERP systems, the applications are coded to make heavy use of temporary files rather than permanent ones because the overhead of making an object permanent on a system is heavy.
And so, IBM has created a temporary file system - technically a user-defined file system with temporary rather than permanent data attributes - within IFS. And in early tests done at IBM's Rochester Labs, the transaction throughput on some ERP systems has increased by between 70 and 80 per cent. The thing with this temporary file system is that the data stored in it will not survive a dismount or an IPL. You do that, and it is gone. This temporary file system can only be created in the System Auxiliary Storage Pool, by the way.
The i 6.1.1 rev also allows for a system to be upgraded from an image stored out on the network rather than from a local boot device on that system.
The i 6.1.1 update is supported Power Systems 520, 550, 560, 570, and 595 machines; System i 515, 520, 525, 550, 570, 595 machines; iSeries 800, 810, 825, 870, 890 machines; and BladeCenter JS12, JS22, JS23, and JS43 blade servers. Customers can upgrade directly from OS/400 V5R3, i5/OS V5R4, and i 6.1 to i 6.1.1, and the upgrade is free to customers who have software maintenance contracts if they have i 6.1 already. It is not clear what the upgrade fees are from prior releases to move to i 6.1.1. ®