The Four Hundred

So many phenomena in life are cyclical. As far as I can tell, this is the third or fourth wave of server consolidation that has swept the system/server market, which has spanned more than four decades. Distributed computing and server consolidation, the two main opposing forces in the data center, may have had different names in the past, but the economics of small and large machines have always encouraged IT shops to evaluate and often change their systems and application architectures.

It's a wonder that IT managers, financial executives, and CEOs and company owners don't get annoyed that servers, applications, and their associated peripherals are always in a state of flux. It must be tempting to just turn the blasted machines off and go play a round of golf. (I know I feel that way some days.) But change, like the distributed computing-server consolidation cycle, is just an inevitable consequence of the particular economics of specific technologies at one point in time. When the economy is weak, as it is now, company managers are looking for ways to save money, because they don't enjoy firing people, and the data center, regardless of the increasing importance of IT to the day-to-day running of just about every business, is always a good place to start to try to save some money, by investing in new technologies and tossing out old ones.

In 2002, every server vendor is talking about the benefits of server consolidation, in terms of lowering the total cost of ownership and administration for print, file, Web, e-mail, and application server workloads. It doesn't matter if they are selling iSeries, Unix, Windows, Linux, or mainframe servers. The advent of sophisticated workload management functions within just about all operating systems--features that used to only be available on OS/400 or mainframe operating systems--and the availability of logical partitioning on all of these platforms has enabled server vendors to make the case that now is the time to take all those servers sprawled all over the place--in the data center, in departments around the office complex, and in remote locations linked over private networks or the Internet--and consolidate them, in an inverse Big Bang, into as few servers as possible, while supporting the exact same clients and applications.

In a conference call with resellers and other business partners last week that made the case for server consolidation onto the iSeries, the iSeries marketing team presented a little more detail than we are used to seeing about how OS/400 shops are using other servers and the economics of consolidating them.

IBM started the call off by reminding everyone that it has offered the Portable Application Solutions Environment (PASE) within OS/400 for two years and that this AIX runtime environment has now been updated to support 32-bit and 64-bit applications compiled for IBM's latest Unix, AIX 5L 5.1. IBM isn't just selling PASE as an AIX consolidation and application porting environment, it is using it: Tivoli Storage Manager and the TCP/IP networking stack in OS/400 are actually AIX programs running in PASE.

At OS/400 shops, however, Unix is not as common as Windows, so Windows is where the server consolidation action is. (This explains, in part, why IBM will take about two years to deliver native AIX support within iSeries logical partitions.) Anywhere from 70 to 80 percent of OS/400 shops have Windows NT or Windows 2000 servers installed. These Windows machines are usually doing print and file serving, but may also be used for Web serving for corporate intranets or Web sites or as application and database servers. A couple of different factors, IBM says, will make Windows server consolidation onto the iSeries directly or onto xSeries servers coupled to the iSeries an attractive option this year and into next year.

First, Windows NT 4.0 will not be available in the reseller channel beginning on July 1, and on January 1, 2003, Microsoft will start charging for support services on Windows NT 4.0 that up until now have been free, such as hot fixes to Windows not related to security. Both Microsoft and IBM say that customers should be prepared to upgrade to Windows 2000, and that they should start looking at it soon. Moreover, Upgrade Advantage, an upgrade protection program from Microsoft that is roughly analogous to IBM's Software Subscription upgrade and pricing scheme for OS/400 and related programs, ends July 31, and the replacement program, called Software Assurance, is more costly, according to IBM. IBM adds that Windows 2000 server licensing is different from Windows NT licensing, and that customers may need to buy more Client Access Licenses (CALs) from Microsoft (not to be confused with IBM's Client Access family of host connectivity software for OS/400 servers), and perhaps even adjunct server programs, to get the same level of functionality and connectivity they have with their Windows NT servers when they move to the server editions of Windows 2000. This could add up to a lot of dough--exactly how much depends on the number of users, the kinds of programs used on the Windows platforms, and the number of servers and processors supporting Windows.

This Windows NT-to-Windows 2000 upgrade cycle, which is just beginning in earnest, is going to affect anywhere from 175,000 to 200,000 OS/400 shops. By IBM's internal statistics, 57 percent of OS/400 shops have from one to five Windows servers, 14 percent have from 6 to 10 Windows servers, 10 percent have from 11 to 20 Windows servers, 17 percent have from 21 to 200 Windows servers, and two percent have 201 or more Windows servers. These statistics are for companies with between 100 and 1000 employees. If, for the sake of argument, you assume that the average site in each one of those bands has a number of servers in the middle of each range (excepting the last one), and then multiply those numbers across the OS/400 base, you get a gargantuan number of Windows servers. Depending on how you wiggle the estimates, you get anywhere from 4.5 million to 5.5 million Windows servers. If you then adjust this to take into account that smaller companies may not have as many servers as those with between 100 and 1000 employees, you are still safely in the ballpark of several million Intel-based servers, which is one of the largest installed bases of such machines in the world--bar none. (These are not all IBM machines, of course. OS/400 shops buy machines from all the major vendors, and some white box suppliers, too.) This base represents billions, if not tens of billions, of dollars of potential Intel-based server sales over the next several years.

The hugeness of that number, the desire of customers to spend less money on servers, and the essentially forced upgrade from Windows NT to Windows 2000 has IBM and its resellers rubbing their hands together with glee. So does the flexibility of the modern iSeries platform. The iSeries can support Windows workloads either through the Integrated xSeries Server (IxS) coprocessor card or through real xSeries servers attached to OS/400, and its file systems and disk storage through the Integrated xSeries Adapter (IxA); through Linux partitions equipped with open source print, file, Web, e-mail, and database programs; or through equivalent software that runs natively on OS/400 or within PASE.

An OS/400 shop, in the example given by IBM to its sales reps and resellers, might have 10 Windows servers: a couple of file servers, a couple of database servers, a couple of application servers, a test server, and multiple Web infrastructure servers (running a firewall, a domain name server, and an e-mail server). This shop knows it has to eventually move to Windows 2000 or to an alternative platform that can support those workloads. It is not thrilled about giving Microsoft more money. Moreover, this Windows network is difficult to administer, and backing up the data stored on it is a nightmare. Some machines are old, some are new, and not all of them will support Windows 2000 very well. Most importantly, all of the servers are not running close to efficiently--the server vendors I talk to say that the typical infrastructure server is running at 10 to 25 percent of processor capacity most of the time.

IBM's example offered two scenarios to migrate this Windows NT network forward. The first is to install five new servers to support Windows 2000 running applications, as well as a new file server, Active Directory server, and a messaging server; the firewall, DNS, and test servers remained in place with only upgraded Windows 2000 licenses. In IBM's example, which used servers and storage from Compaq, these machines share Compaq's Modular Array storage, which simplified backups somewhat. The three DL360 servers, two DL580 servers, and Modular Array with 408 GB of disk capacity cost $103,126; SANworks virtual replication software, VERITAS Backup Exec, and upgrades to Windows 2000 (including Upgrade Advantage and Software Assurance) brought the price up to $149,093.

In scenario two, which is an iSeries server consolidation proposal, this customer, which already has an iSeries you'll remember, is shown instead that it can spend less money by buying two xSeries 360 servers connected to the iSeries by the IxA cards to support the SQL Server database and its applications, and then consolidate the Active Directory, messaging, and testing workloads onto three IxS cards; the file serving, DNS, and firewall applications are moved to Linux partitions on the Model 820. IBM's example assumes that this iSeries customer was already considering an upgrade from an earlier AS/400 platform and is in the processor of buying a suitably powerful Model 820 with lots of disk capacity. In any event, those xSeries servers, IxS cards, and IxA adapters cost $90,342. SuSE Linux 7.0 for the iSeries costs $3,549, including one year of support. Upgrading all of those Windows NT servers brings the total cost up to $106,538.

This is obviously a lot less money. But OS/400 shops had to already be willing to pay a premium to upgrade to the iSeries Model 820 server, where processor, memory, and disk capacity are priced at a big premium, compared with prices for equivalent hardware on the Intel platform. The economics of such server consolidation scenarios will depend in large measure on the efficiency of how those resources are used. An iSeries server processor, for instance, might cost 10 times the price of an Intel processor of equivalent power. But the iSeries would allow a logical partition to be given 10 percent of that processor to, for instance, run a firewall, and that partition could be scaled to run at near peak efficiency.

Of course, with the advent of partitioning software from VMware, whose GX Server and EX Server partitioning programs are sold on IBM's xSeries machines running Windows and Linux, the efficiencies are possible, in theory, using partitioning on those platforms. OS/400 shops, IBM must be betting, will prefer to consolidate as much as possible on the iSeries--with their trusted OS/400 in control of the whole shebang--than to support what amounts to two different partitionable environments--one running OS/400, the other running Windows and/or Linux. This is probably a safe bet for a lot of OS/400 shops.