Database Server/400, Anyone?
February 9, 2009 Timothy Prickett Morgan
The dividing line between a server and a disk array not only blurs from time to time. It also sometimes moves around. A system, of course, includes various servers as well as storage arrays, and the lines are less important. The AS/400 and its progeny have always been systems, even though these systems have been injected with lots of foreign technology–various file systems, runtime environments, emulation environments, and programming languages. Even databases.
The AS/400 was, first and foremost, a database engine, and one with an integrated set of tools for storing information in and extracting it from that nameless database that the AS/400 faithful still call DB2/400. The database was, in fact, the one and only file system in the original AS/400s, and it wasn’t until V3R6 back in the mid-1990s, when IBM was making the transition to 64-bit PowerPC chips, that IBM ripped the database up out of the OS/400 microcode and slapped it down onto the OS/2 High Performance File System that the database stopped being the one and only file systems on AS/400s. This was, perhaps, a bad idea for the long haul. But it made client/server and Web computing run better in the short term, and has allowed the AS/400’s offshoots to support technologies originally developed for Unix, Windows, and now Linux platforms.
But despite all of this, even today’s Power Systems i box is still mostly a database engine. A number of times in the past, I have suggested how Big Blue might take advantage of the low-priced economics of the entry AS/400 and iSeries platforms and the integrated database clustering technologies in DB2/400 to take on X86 and then X64 platforms on their own turf and maybe steal a little business. (See the Related Stories section at the end of this story for more on those riffs on the possibilities.) But there is, of course, another alternative. And if the sketchy rumors I am hearing turn out to be true, IBM might just be thinking of using the brains inside of its high-end DS8000 servers to run instances of DB2 databases and act like adjunct processors for the DB2 databases running on Power Systems servers.
The distinction between a Power Systems server and a DS8000 (and for that matter, other DS series disk arrays) is not so much technical as it is economic and nomenclature. Since IBM shipped its first “Tarpon” disk arrays back in the late 1990s, followed by several generations of “Shark” arrays that culminate in the current TotalStorage lineup from IBM. Starting with Tarpon, IBM took Power-based servers and attached lots of disk drives to them and started adding storage software to a stripped down AIX instance running on the box. Over time, this software has grown to be more sophisticated, allowing for flash copying of data sets to boost performance, data replication between mirrored disk arrays for high availability, and all sorts of wizardry to make file systems more resilient and flexible.
Imagine, if you will, if a cheap copy of the DB2/400 database was plunked onto a DS8000 disk array. Now imaging that the data in your production system was replicated on the fly to this machine, which we might call a DB2 data server. A database license on this box might cost $13,000 per processor, a quarter of what it costs on a core on a real Power Systems machine for high-end boxes. Now, IBM would not let you run your production applications off this copy of the database, of course, any more than customers using a System z Integrated Information Processor, or zIIP engine, to accelerate DB2 workloads can run their ERP applications off that zIIP. The zIIP exists to make DB2 run faster in a more economical way. And putting DB2 UDB and DB2/400 on DS6000 and DS8000 arrays can accomplish the same exact feat for Power Systems shops. The DB2 instances running on the DS disk arrays would do some of the SQL and native query processing and then feed data back to applications as if it had come from the production database, which would be busy, of course, doing useful work like processing business transactions.
It would not surprise me at all to see a pairing between the Power 595 and a specialty DB2 and DB2/400 engine based on the DS8000 arrays, because IBM’s largest customers are familiar with the specialty zIIP and zAAP (for accelerating WebSphere and Java workloads) engines on mainframes. The Power Systems users want the same kind of discounting as mainframe shops are getting, and they have the same complex queries and data warehousing workloads that the zIIP is aimed at. A zIIP processor core costs roughly a quarter of a regular core that is turned on to run z/OS, and it is not unreasonable to expect the same kind of economics with Power Systems.
So why not just designate the equivalent of zIIP and zAAP cores on Power Systems directly and leave everything on the server? Two words: Oracle and EMC. If the database acceleration function is on the server, then IBM has to extend the functionality to Oracle databases on Power Systems, and it has to cut some sort of deal to get discounted pricing for Oracle 11g for its Power Systems customers if they choose the DS8000 as a database server. Oracle doesn’t want to do this. And neither does IBM, really, but global economics may force it to be clever and come up with ways to provide more bang for the buck for server customers. By putting the database acceleration function inside the disk array, IBM can use the Transformation Server data replication technology it got through its DataMirror acquisition in September 2007 to have data in Oracle or other databases replicated to DB2 instances on the DS8000 array. It can accelerate Oracle database queries–or take some of the load entirely and pass back answers directly to select applications–and deny Oracle a sale. The other reason to do it this way is to try to take some sales away from EMC, which doesn’t have such data replication facilities–but can either build it or buy it if the idea has economic merit.
While what I am hearing about is the possibility of a DS8000-based DB2/400 and DB2 UDB database server, IBM could just as easily create database servers on entry DS3400 and midrange DS5000 arrays as well for accelerating database workloads on entry and midrange Power Systems boxes. Call it the Database Server/400, just for fun. I have not heard anything solid, just IBM and some users kicking around the idea. Let me know what you think of the idea. I’m interested.