A reason for Unix sysadmins to thank Microsoft

Specifically, an IDC study on the choices people make when considering Unix/Linux or Windows as a migration target is worth reading on its own.

The paper outlines part of an analysis of 400 interviews done in 2005 with people somewhere in the decision making chain in companies that are considering migration from a Unix such as HP-UX, Tru64, AIX, SCO, Solaris, or Linux to Windows or another Unix.

IDC's key conclusions were:

• In the Unix/RISC world, Windows did best against IBM AIX and HP-UX, as surfaced in the study results, while Linux did best in replacing Solaris. IDC believes that ISV packaged application availability may help to explain Windows affinity for the AIX and HP-UX customers.

   

• "In the study sample, Solaris was seen to be far more impacted by migration to Linux, although Solaris was less impacted by competing Unix operating systems (12%). IDC believes that a large part of this trend is driven by IT and Web infrastructure workloads deployed on Solaris, which move more easily to Linux than more complex and demanding enterprise workloads (e.g., BI, OLTP, and other business processing applications).

   

• "Although 42% of all IBM AIX migration in the study sample was to Windows and 22% was to Linux, 15% of IBM AIX migration was to Solaris and HP-UX, with Sun Solaris the bigger platform winner.

   

• "While HP-UX was more exposed to Linux migration than AIX (34%), it was also challenged by Windows (40%) and other Unix variants (19%). For cases in which HP-UX migration moved to other brands of Unix, IBM AIX appeared to win more share than Sun.

   

• "By a strong 4 to 1 margin, SCO Unix users reported that they preferred Windows as their new server platform. IDC believes this preference may be driven by SCO's ISV and reseller community as well as the fact that SCO Unix has always been deployed on an x86 server platform, rather than on a RISC platform.

The most interesting results given are those pertaining to the IBM and HP communities - the former because nobody chooses AIX on product merit and the latter because most of the people still using HP-UX and Tru64 continue to do so mainly to prove they were right not to choose Sun in the ninties. For example, 56% of sites reported some AIX use on a similar survey done in 2003, while 55% do so today - and compare that stability with the disaffection of HP's base where a combined 81% showing for HP-UX or Tru64 in 2003 dropped to only 58% in 2005.

Solaris, incidently, went up over the period: from 71% to 80%.

Notice, however, that Linux numbers aren't reported - and that's because it's to Microsoft's advantage to support the idea that Linux isn't Unix. This report, for example, could just as well have appeared on Red Hat's site as Microsoft's; an oddity explained by the reality that Microsoft sees Lintel as a kind of halfway house, a mere stop on the road from Unix to Windows, and therefore supports Red Hat's anti-Sun campaign exactly as they would the activities of any other strategic partner.

It's important, in this context, to note that Microsoft's support for Linux is ultimately driven by its contempt for the people who day in day out sell Microsoft to their employers - because those are the people who take over from the Unix experts when Lintel replaces Solaris, HP-UX, or even AIX and whose expertise then drives the downstream slide to Windows.

A cynic would notice, too, that the study's most dubious result is reported with respect to Solaris - Figure 4, page 7, consists of a barchart showing less than 20% of 835 Solaris applications being considered for migration - but Figure 6, page 10, is a barchart showing that more than 50% of 341 Solaris users are considering migration to Linux.

Beyond that, however, there's a reason that Microsoft would be well advised to get this study off their "get the facts" site - and why Red Hat shouldn't then pick it up. The reason is that recent hardware change makes the explicit message: that web server applications should should be migrated to Lintel, a strategic mistake for both of them.

That hardware change is the release of Sun's Niagara One, or T1 UltraSPARC - the first product in the CMT/SMP "coolthreads" line to hit the market.

Prior to the T1, the case for moving a web application from something like a Sun 450 bought in 1999 to a new Lintel box was pretty good - no significant application change, similar or better performance, 25% of the power cost, 10% of the space and maintenance cost- and really only went wrong for customers who paid Red Hat's fees and/or put Wintel people in charge.

What the T1 does is reverse that cost case by making the argument for moving web applications from xtel to Solaris/T1 overwhelming: up to ten times the throughput with half the input cost - and, incidently, creating opportunities for Unix people to take back operational control.

It's not just web applications either: any multi-threaded workload in which neither floating point nor absolute minimum execution times are critical requirements is a candidate. Consider, for example, this report of a four core, 1Ghz T2000 blowing away a pair of 2.4Ghz Xeons with hyperthreading by a factor of about 2.7 on an Erlang benchmark with 16 concurrent threads - and the next generation will handle 64 concurrent threads, be competitive on floating point, and still use the same rack, the same power, and the same software.

In other words any Unix sysadmin who reads this Microsoft sponsored study should see it as job insurance. Just download and distribute the study, ensure that the bosses understand the cost case Microsoft's consultants are making for migrating web applications to Linux, and then show them the T1 numbers -because Microsoft's logic is right, and any web applications on site should be immediate candidates for migration to the T1/Solaris combination