IBM Takes System i Disk Clustering Up a Notch with HASM
February 4, 2008 Timothy Prickett Morgan
There are a lot of ways of skinning the high availability and disaster recovery cats in the System i market. And the merger of the System i and System p divisions last year into a single Power Systems division, which creates the Power server platform, its operating systems, and high-level systems software, is culminating in yet another product coming from IBM to address the HA and DR needs of System i shops: High Availability Solutions Manager, which was announced with i5/OS V6R1 last week.
According to Steve Finnes, business continuity product offering manager for the Power Systems division, HASM is a “first cousin” of another IBM clustering product created by the company for its AIX-based server line, called HACMP. (That’s short for High Availability Cluster Multi Processing, which IBM launched in 1990 for its RS/6000 server line based on a set of code created by an independent HA specialist called Availant. i5/OS and OS/400 HA software maker Lakeview Technology acquired Availant in 2003 by, and then was itself bought by HA software rival Vision Solutions last summer.) HACMP is not just a two-node clustering solution, but scales up to a maximum of 32 nodes, which can be used as standbys or to do actual work. The new HASM product, says Finnes, was developed in IBM’s Rochester labs with the cooperation of the Austin AIX team that codes the HACMP product. HASM is based on the same basic clustering concepts that HACMP employs, such as cluster resource groups, but AIX has segmented memory and a different file structure, which means HASM has to be specifically tailors for the single level storage (main memory and disk virtualized as a single address space) and i5/OS file concepts. Finnes says that the HASM product can also be thought of as a kicker to the Copy Services for System i Toolkit, which was created by an IBM system programmer named Selwyn Dickey of the Rochester labs last year for allowing System i shops to configure and setup Global Mirrors (asynchronous links for replicating data between two external DS8000 arrays that are separated by large distances) and to instigate roll swapping between clustered servers attached to those DS8000s.
IBM’s decision to put together HASM is something that has been driven by the largest AS/400, iSeries, and System i shops, explains Ian Jarman, manager of Power Systems software, a new units within the Power Systems division that brings together a number of high availability software products, including HACMP, HASM, and the iCluster products from DataMirror, which IBM acquired last fall. (DataMirror was a rival of Vision Solutions, Lakeview, and iTera, all of which have been merged into Vision Solutions.)
“There have been many customers who have been looking to us for the past 10 years to support a disk clustering solution that is scalable and flexible enough to be deployed with storage area networking technology in the data center,” says Jarman. “This is absolutely connected to a broader move to SANs among our largest customers. These companies have deployed SANs for their other server environments, and they want the System i to be part of that same SAN mainstream.”
The advent of HASM doesn’t mean that there is no place for the logical replication high availability software products now sold by IBM through its DataMirror product or those made by Vision Solutions and Maximum Availability, which are based on the remote journaling technology that was woven into OS/400 and i5/OS years ago. These are complimentary technologies. For instance, companies may use HASM and its switched disk approach for campus or metro-level synchronous data replication and the related Cross-Site Mirroring (XSM) software to do role swapping between source and target machines in a cluster, and then they may use a logical replication program such as MIMIX, Echo2, ORION, or NoMax to make a hot backup of the production machine at a remote location for disaster recovery purposes.
Perhaps equally important, while it is possible to deploy HASM on internal disk arrays inside iSeries and System i boxes, it requires the new i5/OS V6R1 operating system (which none of the logical replication products do) and it is most flexibly deployed with IBM’s midrange DS6000 and high-end DS8000 series of disk arrays, which have all kinds of replication functions built into them that work in conjunction with HASM. Moreover, for many customers, particularly smaller shops with modest HA needs, deploying XSM and independent auxiliary storage pools (iASPs) is a less costly and less complex approach than using the full-blown HASM product; this requires OS/400 V5R3 or i5/OS V5R4, which means companies on earlier releases can deploy only XSM for local disk clustering. For such customers still on earlier OS/400 releases, logical replication products that have been the HA default for nearly two decades are still the only game in town. (Another reason to keep current: You have more options.)
The HASM product, which goes by the 5761-HAS feature code in the IBM catalog, is going to be available on March 21, just like the i5/OS V6R1 operating system it requires. The software requires a feature of the i5/OS operating system called High Availability Switchable Resources, which is Option 41, which ranges in price from $2,085 on a P05 tiered System i machine to $62,000 on a P60-class machine. IBM is charging the same price for HASM on i5/OS as it charges for HACMP on AIX boxes: $3,200 per processor core. (You have to add these two charges together to get the cost of the overall HASM solution.) Here’s the interesting bits about this pricing. First, it is a lot less expensive than HACMP with the equivalent Extended Distance (XD) and Smart Assist (SA features). And the HASM stack can also be less expensive than IBM’s own newly acquired iCluster software, which does logical replication, not disk-based replication. Take a look at this table that IBM made for customers that shows this:
Of course, you have to buy disk arrays for certain HASM functions, and IBM charges for its Metro Mirror, Global Mirror, and Copy Services in the DS6000 and DS8000 arrays based on the capacity of the disk array. So these are additional costs above and beyond the software comparison above. (Mirroring functions in the DS8000 array range from $16,272 for a 1 TB capacity up to over $1 million for a 200 TB capacity, and Copy Services is about three-quarters of this price.)
IBM is not supporting the new HASM software on every iSeries and System i box. It can be used on the new JS22 Power6-based blade server (which requires V6R1 anyway), as well as iSeries and System i 520, 550, 570, and 595 machines; System i 515, 525, and 570 machines; and iSeries 800, 810, 825, 870, and 890 machines.