By Timothy Prickett Morgan
Sibling rivalry among IBM’s strategic product lines has always been keen, and when IBM ships OS/400 Version 4 Release 4 (V4R4) on February 26, the operating system for AS/400s will include most of the features that make mainframes and high-end Unix servers worth buying as opposed to AS/400s. The advanced software that IBM has put into OS/400 is not just there to give IBM’s AS/400 sales reps something to talk about. Increasingly, software is the key differentiator when it comes to midrange server sales. Several of the AS/400’s strong points, scalability and reliability being the big ones, are less key factors than they have been in the past for AS/400 customers long in the product line as well as for customers new to the base (about 29% of AS/400 sales in 1998 went into new customer sites, which isn’t bad for IBM). Scalability is less of an issue since AS/400s can have up to twelve of IBM’s very powerful 262-MHz Northstar 64-bit PowerPC processors and have more than 300 times as much online transaction processing capacity as the original high-end B60 systems announced in June 1988. This gives customers plenty of headroom, but not many customers come even close to needing such a large machine. (Server consolidation could change that, of course.) The point is, with the Northstar servers and OS/400 V4R4, AS/400s have more scalability than most customers with several thousand employees or less will ever need, which is the same as can be said for mainframes and Unix servers, and when Intel and its partners get eight-way Pentium II Xeon servers out the door later this year, it will be true for PC servers as well. As far as most server buyers are concerned, AS/400 hardware is not much different from mainframe or Unix or PC server hardware. This has not been the case in the past, but will be true from now on. A vendor may have a three- to six-month lead, but will never be able to hold it since processing power increases are mostly made through advances in microprocessor technology, which follows a predictable curve. Reliability, another strong suit for the AS/400, is less of an issue for AS/400 customers as well as those choosing alternative platforms. Compared to only a year or two ago, midrange hardware is so inexpensive that it has become practical to double up processors and use software mirroring techniques to get high reliability for strategic production applications.
Must have an edge
During 1999 and well into the next millennium, server companies that can show they have an edge in price/performance, total cost of ownership comparisons or software functionality will be able to sell more machines; those who do well in more than one of these categories will fare better. The AS/400 group, as one might expect, will focus on TCO and software functionality. First and foremost, OS/400 V4R4 will become the first release of the AS/400’s operating system that allows SMP-capable AS/400e servers to take advantage of logical partitioning, which up until now has been a mainframe-class function that is extremely useful in big data centers that have to support diverse workloads and have to do so on a budget. The logical partitioning that comes in OS/400 V4R4 (which has more than twice the code of OS/400 V4R3 thanks in large measure to this feature and universal database extensions to the AS/400’s integrated relational data base management system, DB2/400) is similar in concept to the mainframe logical partitioning that is available to customers using IBM’s flagship OS/390 operating system, but it is implemented quite differently. Under V4R4, an AS/400 can have a maximum of twelve processor partitions (one for each CPU in a twelve-way Northstar server). Each partition, which runs its own copy of OS/400 and has its own system clock, has its own dedicated memory and disk I/O resources. Memory increments can be as small as 1 Mb and dedicated disk can be as small as one SCSI controller. There are high-speed connections between these partitions across the AS/400’s internal crossbar switch. In future OS/400 releases, probably OS/400 V4R5 with the Pulsar servers due in early 2000, dynamic reallocation of resources will be possible, but right now partitions are set with software and require a machine reboot to take effect. Under OS/390, a single server can have up to 15 partitions spread over 12 processors (IBM and Amdahl only sell ten-ways, but Hitachi has a twelve-way OS/390 server using IBM’s G5 CMOS engines). Memory in a mainframe logical partition can be sliced anywhere as small as 1 Mb to 4 Mb per partition, depending on the vintage and make of the mainframe, and I/O is allocated at the disk device level. Both OS/400 and OS/390 have much more sophisticated partitioning than that available for Sun Starfire servers, which have 64 processors max and use physical hardware partitioning rather than logical OS-level partitioning; early mainframe partitions worked in the same manner. The Solaris partitions are done in four-way increments and only a maximum of eight partitions are supported on a single 64-way box. Memory increments are 512 Mb, which isn’t very granular, and I/O is only partitioned across two buses. Resources can be reallocated dynamically on a Starfire, but for them to take effect, data bases have to be shut down and restarted. Hewlett-Packard’s 9000 servers and their associated HP-UX operating system do not offer either logical or physical partitioning.
So what’s the big deal about partitioning? Server consolidation, for one thing. Without partitioning, there is no way to have server consolidation for European and multinational corporations. While the AS/400 has always supported multiple, simultaneous languages on a single server for OS/400 and application screens, it has not been possible to have two different system and date clocks to truly isolate jobs running in different time zones and countries from each other on a single machine. Moreover, for service bureaus using AS/400s, partitioning will allow simple cost accounting of IT resources that has made the services business relatively easy to support on mainframes and very difficult to do on other platforms. (Remember, it is always cheaper to buy one big server than two little ones of equal power, and this is especially true of IBM mainframes and AS/400s. So the only way to make money in the services business is to use gargantuan machines and slice them up.) But partitioning works on smaller SMP servers (except those in the model 170 Invader line, which doesn’t have the I/O to support it) and will likely also be useful for customers who want to have a test partition to develop applications on the same machine that actually runs their production applications. It’s a lot easier to move data from one partition to another for testing than it is to set it up on a whole new machine. And when you’re through testing in a partition, the resources used are available to support other real work. Partitioning isn’t the only new software functionality to be included in OS/400 V4R4. IBM has been working with high availability software vendor Vision Solutions to create extensions to OS/400 that put sophisticated clustering right into the operating system and very close to the iron. Prior to V4R4, clustering software from Vision Solutions or its competitors in the midrange HA market, Lakeview Technology or DataMirror, ran like any other system or application program, up on top of OS/400. But in V4R4, fault tolerant clustering APIs are down at the machine level in microcode, which can be accessed by third party applications from these three vendors, and presumably, others when they enter the market to make existing Unix and NT clustering schemes link into AS/400s. The continuous availability clustering in V4R4, as it is being called, provides a single resource view of multiple machines in a cluster. This cluster can have from 2 to 128 nodes, which have a single resource view to management programs but, for now, still look like separate machines to most applications aside from simple IP web serving. V4R4 cluster nodes can provide failover backup, but by 2000, says IBM, many applications will be enabled (with the help of partners like Vision Solutions, Lakeview and DataMirror) for full Tandem- style, load-balanced clustering across those nodes. Clusters can also be created between logical partitions in a single machine for failover as well.
As machines grow, it gets increasingly difficult to backup large data bases, and that is why techies at the Rochester Labs have developed a parallel tape backup function that allows users to back up large data bases by running up to 32 different backup jobs simultaneously to archive a data base on multiple tapes. To give a feel of how good this works, consider that it took a first generation RISC AS/400 four-way 53S server running V4R1 over 40 hours to backup a 1 TB data warehouse in August 1997. Last year, running OS/400 V4R3 on a twelve-way Northstar server, it took 20 hours to do the same job. With parallel tape backup on the same machine with V4R4, the 1 TB data warehouse could be archived in 66 minutes. This is a very big improvement. The other big change in V4R4 is object relational or universal database support. The AS/400’s integrated data base, DB2/400, is being given full support for binary large objects (BLOBs) like other universal data bases have long since been given. That’s why DB2/400 will henceforth be called Universal Database for AS/400 (yet another unimaginative IBM name for a product). Customers wanting to integrate spreadsheets, video clips, still images, sound bites or other binary files into their relational data bases can now do so, and they can do it in two ways. First, they can literally plunk a file, up to 15 Mb in size, right into the database table and leave it there. If they do that, however, it could impact online transaction performance and make archiving a real nightmare. Alternatively, customers can use the DataLinks feature, which stores BLOBs outside of the data base and just keeps pointers to these files inside the data base. This, of course, complicates the backup job, since there is no way, as yet, to tell OS/400 to archive a data base and its associated BLOBs. It has to be done manually. The good news is that OS/400 won’t let anyone accidentally or purposefully delete a DataLinked BLOB from a machine until its link is removed from the data base. Universal database support for V4R4 won’t be ready until September, and IBM isn’t sure how it will distribute it to customers yet. It will likely be done with a PTF, IBM-speak for service patch.