There’s no question in anyone’s mind that the proliferation of servers in corporate computing has created an unprecedented and rich mix of applications and connectivity for users. But distributed heterogeneous computing, where anywhere from two to dozens of different operating environments are scattered around dozens to thousands of servers, has also created a nearly unmanageable mess.
It comes as no surprise that we are now in the middle of a backlash, at least partially reversing the trend from distributed to consolidated application, print, and file servers. It certainly can be said that the recent advent of server consolidation (and its related phenomenon, storage consolidation) is just one in a long list of opportunistic initiatives put forth by server consolidation’s biggest champion, IBM, to sell more computers to its customers. However, IBM has a point. MIS managers, unable to keep control of deployments of departmental servers and PCs throughout their companies as server and PC prices have plummeted to the point that department managers can buy equipment out of petty cash and bypass the whole MIS department (until problems arise, of course), have got to get control of all the equipment and software scattered hither and yon around their companies. In many cases, that means consolidating servers and storage and putting in higher-speed networks to keep users as connected and seemingly as close to their servers as they were in the past when they were in the next room.
Just how big is server consolidation? It depends on whom you ask and how you ask the question. By IBM’s estimation, server consolidation is about a $2 billion potential business with about a 40 percent growth rate. In a recent poll of 250 customers by IBM, 60 percent told Big Blue that they were considering server consolidation.
A survey of large enterprises by SoundView Financial Group, Inc. found that 88 percent of the respondents were, in fact, consolidating servers already. An InformationWeek poll found that 72 percent of the companies polled are recentralizing distributed systems and that 20 percent had already done so. A survey by The Research Board found that 80 percent of multinational companies are recentralizing their information technology (IT) resources. The Conference Board says that 68 percent of multinational companies are recentralizing their finance and accounting systems. A Computer Economics
poll found that 81 percent of the companies that consolidated their servers had significant cost savings, 44 percent got better manageability and availability, and 41 percent got improved performance. Clearly, there are some strong drivers behind server consolidation.
It’s Not Just a Mainframe Thing
Even though IBM’s initial focus for server consolidation was for S/390 customers, server consolidation is equally valid when applied to AS/400 customers. While IBM has not offered up statistics for the AS/400 base, the typical AS/400 shop has three to four other LAN servers at its sites. There are easily millions of non-AS/400 servers at AS/400 sites, many of them PC servers running OS/2, NetWare, and Windows NT. The average AS/400 customer also has two or three AS/400s, many of them in remote sites.
It would be difficult to find an AS/400 shop that doesn’t also have a slew of PC or UNIX servers that perform myriad jobs not suitable for AS/400s. In the past, AS/400s were not very good at simple print and file serving for PC clients, so customers usually have investments and skills in true network operating systems (especially Novell’s NetWare and IBM’s OS/2) that often go back as far as their investments in AS/400s. While print and file serving can be moved to Integrated PC Servers (IPCSs), IBM has just killed everything but Windows NT for future machines, thus making the IPCS an unattractive alternative for the large installed base of NetWare and OS/2 customers at AS/400 sites.
Other servers at AS/400 sites run applications that cannot be run on the machines, the most prominent being UNIX applications such as computer-aided design (CAD) and electronic design automation (EDA) applications—commonly used to design products at manufacturing companies—as well as geographic information systems (GIS) used by telecommunication, transportation, and utility companies. Suffice it to say, AS/400 shops are stuck, by history and by necessity, in supporting anywhere from one or two to dozens of non-AS/400 servers at their sites. And this makes support costs skyrocket.
When IBM talks about server consolidation in the AS/400 base, it is talking about a couple of different things. First, there is the obvious consolidation of multiple, distributed AS/400 systems with single, large, symmetric multiprocessing (SMP) AS/400s. This kind of consolidation has really only been possible since the AS/400e Apache servers were introduced in the summer of 1997. Only then could IBM deliver eight-way and 12-way AS/400 systems and servers that were capable of supporting the large workloads running on several of its existing four-way E, F, CISC AS/400, and RISC AS/400 servers and still leave customers with some headroom as their workloads continued to grow.
Last year’s high-end Northstar AS/400e models offer almost twice again the throughput capability of the Apache models, which means that even customers with big distributed Apache servers can consolidate their workloads for potential savings. And when the Pulsar AS/400s come out in the middle this year, with about twice the aggregate throughput of the Northstars, customers at the high end will again be able to consolidate workloads and servers onto big SMP servers.
The advent of hardware partitioning, widely expected with Pulsar systems and OS/400 V4R4 later this year, will allow a more sophisticated kind of server consolidation. One of the primary concerns with server consolidation is that, as more applications are added to a single machine, tuning that machine to optimally support those applications simultaneously becomes a very tricky job. If someone starts a complex query at the same time as a spike in order-entry activity occurs, for instance, everybody is affected adversely.
But with V4R4 servers, IBM plans to add the kind of hardware partitioning that has been the hallmark of its mainframe systems for the past decade. Basically, partitioning allows a big AS/400 to be chopped up into multiple virtual and isolated AS/400s for the purpose of running applications. Rumor has it that processors and blocks of memory, as well as associated disk controllers and subsystems, will be able to be linked in a partition, which greatly simplifies application tuning. The good bit about partitioning is that an AS/400 can be rejiggered as workloads and conditions change.
Server Consolidation Is Often Software Consolidation As
Well
AS/400 server consolidation also means bringing workloads that have been running on non-AS/400 servers back under the skins of the AS/400. There are a number of workloads running on PC servers that IBM feels are better served from the AS/400. Lotus Domino, which has been native on the AS/400 for almost a year now, is perhaps IBM’s best example. Thousands of customers were running Lotus Notes and Domino on the AS/400’s IPCS, and thousands more were likely running it on outboard PC servers from the usual vendors (IBM, Compaq, Dell, Gateway, and a few others). Clearly, IBM has made great strides in moving a big portion of these customers to native implementations of Domino. With 27,000 Domino Mail users supported on a 12-way Northstar server, IBM can make a convincing case for customers to consolidate Domino onto AS/400s. But since Domino groupware and email software was not typically installed at most AS/400 sites before it was announced for the AS/400 itself, IBM can’t really consider Domino sales from this point on as being server consolidation.
Another important job at AS/400 sites that IBM wants to get off of UNIX and NT servers and back onto the big black box in the data center is Web serving. That’s one of the reasons IBM is so gung ho about bringing the Apache Web server to the AS/400. The company sees the Apache Web server not only as a standalone Web server available like the UNIX and NT implementations through the Apache Group, but also as an integral part of its WebSphere Application Server e-commerce application development environment for NT, UNIX, AS/400, and S/390 servers.
By the time you are reading this, IBM will have the first release of the Apache Web server integrated into WebSphere and will have worked with the Apache Group to get the AS/400 version up on its www.apache.org Web site for free download to AS/400 customers. (By this time next year, Apache will probably be bundled with Lotus Domino and OS/400 as well, effectively replacing Internet Connection Server, Domino Go, and HTTP Server for OS/400, the three incarnations of IBM’s native Web server for AS/400s.)
The Apache Web server was the first Web server program to allow multiple Internet domains to be housed on a single physical Web server, and, not surprisingly, this and the fact that it is available free has made it the most popular Web server on the Internet and amongst companies doing business on the Web through secure Web sites. Apache, according to market researcher Netcraft, accounts for over 52 percent of the domains on the Internet; it probably has a lower percentage by the number of physical servers since many big Web-hosting firms can put hundreds of sites on a modestly powerful UNIX server. With the adoption of Apache for the AS/400, IBM lets AS/400 shops tap into the skills base of the Apache community while letting them keep their Web sites back on the AS/400, where most companies feel they belong.
The remaining type of server consolidation that IBM is emphasizing with AS/400 customers is the movement of jobs running on external PC servers to the AS/400’s IPCS. With IBM removing support of NetWare and OS/2 on future IPCS cards and future OS/400 releases, it is a little bit difficult for IBM to make a convincing case for PC server consolidation across the board. But IBM’s continued and future emphasis on Windows NT Server for the IPCS will probably strike a chord with AS/400 shops frustrated by the proliferation of external PC servers running all manner of print and file serving operating systems.
Server consolidation on the IPCS, for most AS/400 shops, will mean not just removing hardware, but also changing print and file serving environments. IBM has undoubtedly decided that NetWare and OS/2 are dead, or else it would continue to support them. Plenty of AS/400 customers are miffed about IBM throwing in so exclusively behind Microsoft, especially since NetWare and OS/2 are widely perceived as being more reliable and better performing than Windows NT Server.
Perhaps more important for the future, IBM has not added support for AIX or Linux to the IPCS, although there is a special AIX version of the IPCS available on a special bid basis if AS/400 shops have a compelling need. That IBM did not promote tight links between its own UNIX and OS/400, especially when it was clear a few years ago that the Internet and its UNIX-based software were going to take over the world, shows just how cynical IBM can be about AS/400 shops and their adoption of new technology. More often than not, it is IBM and its high prices and limited vision in the AS/400 base that is holding up progress.
If Linux takes off, it would be a snap for IBM to support it on the IPCS, but doing so would probably get AS/400 customers putting their Apache Web servers on Linux (Apache on Linux is the most popular combo on the servers that comprise the Internet) rather than putting them on the AS/400. IBM makes a lot more money selling AS/400 processing power than it does selling IPCS processing power—10 to 200 times as much for raw processing power, depending on the AS/400 make and model. So, don’t hold your breath waiting for Linux on the IPCS.
While IBM has not been as open with the IPCS as its earlier pronouncements of the product had promised, the company will nonetheless sell lots of IPCS cards to AS/400 customers who want to better manage their Windows NT print and file serving networks. The NetServer features of OS/400 V4R3, which make the AS/400 look like a Windows 95/98 or Windows NT peer in a Windows peer-to-peer network, will also encourage many AS/400 customers to dump external Windows NT and NetWare PC servers that currently handle print and file serving. AS/400 customers really don’t need IPCS servers, much less external PC servers, for simple print and file serving to Windows clients any more, and they would be well advised to get rid of them if they have the extra capacity to support these jobs on their AS/400s—provided that NetServer performs acceptably.
How Much Savings Are We Talking About?
Server consolidation doesn’t necessarily just mean fewer servers and lower administration costs. It might also mean fewer operating systems. IBM’s server consolidation team leader, Richard Fuchs, contends that the real driver behind server consolidation is software and related services. He says that about half the cost savings driving server consolidation at customer sites come from software and services, with server hardware being only 30 percent and storage (disk and tape subsystems) being another 20 percent.
Those numbers give the basic gist of what to expect, but they are not very specific to the AS/400. However, basic economic factors have always provided an argument for buying one big AS/400 rather than multiple smaller ones. At current street prices, low-end 600 and 620 series AS/400e systems sell for as much as $180 per Commercial Processing Workload (CPW) client/server performance point. (CPW is the benchmark that IBM uses to gauge relative system performance in AS/400s. It replaced RAMP-C a number of years ago and is loosely based on the open TPC-C benchmark.) But bigger 640 and 650 series systems sell for $80 to $100 per CPW, and the biggest 650, the 12-way model 2189 Northstar system, sells for $60 per CPW. In the server side of the AS/400e line, small S10, S20, and 170 Invader servers have street prices that range from $20 to $40 per CPW, but big S40 servers cost as little as $16 per CPW.
Maintenance costs are also considerably cheaper on big AS/400s when tracked against processing power. Small AS/400e systems and servers have monthly maintenance fees of anywhere from $2 to $5 per CPW per month, but big AS/400 systems and servers come in at below $1 per CPW per month. Memory costs per megabyte, however, are twice as expensive on bigger servers than on smaller ones, so some of the processor and maintenance cost savings from server consolidation are eaten up by increased memory costs.
What’s the bottom line? The exact amount of cost savings that customers can expect from AS/400 server consolidation depend in large measure on their particular
circumstances. A recent study by International Data Corporation (IDC) analyzed some very simple AS/400 consolidation scenarios. In one scenario, a customer with five AS/400s in the same location consolidated onto a big AS/400 and realized a 15 percent savings—and still ended up with almost twice the aggregate processing power. In a scenario of a company with 20 distributed AS/400s being centralized into one big machine, the customer added 60 percent more aggregate power and still realized an overall savings of 13 percent. (To check out the IDC report, go to www.as400.ibm.com/conslt/tco.htm.) Obviously, if customers added only 20 percent more aggregate power, they would have realized bigger savings, but most server consolidation strategies make sense only as an overall processor upgrade strategy. It is usually better to add capacity for the future and consolidate all at once.
LATEST COMMENTS
MC Press Online