Having come of age in the IT industry at the likely zenith of the mainframe at the start of the 1980s, I am always drawn to mainframe-oriented news. For the past 15 years, though, such stories have been of the kind in which the writer is somewhat amazed that the mainframe is "not dead yet." After some two decades of having its market share eroded by migration to server-based applications, "big iron" is back. And, irony of ironies, the catalyst for the comeback is the need to deal with server farms that have grown out of control.
Where even very large companies might only have had a dozen or so mainframes (albeit large and expensive), it is not unusual today for quite small companies to have that many servers. For larger companies, the server farms can number in the hundreds or even thousands. That was certainly the case with IBM itself. In early August, IBM announced that it was doing away with some 3900 physical servers and replacing them with some 30 mainframes. According to IBM, the energy consumed will be 80 percent less, while system and software support and licensing will be reduced dramatically.
Another irony — it is far easier to move from server to mainframe than it was to migrate from mainframe to servers in the first place. Back then, mainframe programs had to be rewritten from scratch (generally) to run on servers. Moving back, though is simple since the consolidation is virtual. With virtualisation all the rage today, it is interesting that IBM invented OS virtualisation with its mainframes some 40 years ago.
Today's System z mainframe is not the one that users left in droves over the past two decades. Back then, IBM owned the datacenter. Filled with usually proprietary IBM hardware and software, users had little choice but to accept what IBM offered. Change was slow, and almost anything with the IBM logo on it cost a lot more than it should have.
But the new mainframe can run Linux in addition to IBM's z/OS and, as noted, the key point is virtualisation. Referred to as LPAR — or logical partitioning — IBM has had this hardware-assisted virtualisation for several decades now. Each mainframe can run hundreds of "logical servers" at once.
So all of this is good news for end users running out of space and energy (electrical and human). It is bad news for server vendors, as well as for vendors that price their software by physical CPU — though a licence change can put them back in the money. It will be interesting to see how server vendors react to this.
Because the mainframe is running Linux and communicates via Ethernet and TCP/IP, the staggering learning curve that was part of the IBM experience years ago is gone. What IBM fails to note is what a good cross-over point is -- 50 servers, 500 servers? There are no doubt some "gotchas" and limitations, but with potential benefits as vast as this, big iron is back in the ballgame.