Your experience with recompiles agrees precisely with mine. Something always goes wrong in moving an application across operating systems. But when I made the recompile statement yesterday, I wasn’t thinking about generic server operating systems being worked on by enterprise application programmers. I was thinking about embedded systems development, in which a recompile should be the least of worries, assuming access to source code. (I will admit to the same knee-jerk reaction you had to “mere recompile”, but then I thought a discussion of a Murphy’s Law was an unnecessary digression.)
What I did not point out yesterday was that FreeBSD has a Linux compatibility layer that traps Linux system calls and converts them to FreeBSD system calls. As you know, this means Linux applications can run straight out-of-the box on FreeBSD – without even having to recompile the source code. This must be similar to the library wrapper you describe Solaris provides for Linux executables.
I remember years ago when Sun introduced a Windows emulator for its Unix workstations. At first intrigued, I soon discovered that the best of the emulators was embarrassingly slow and unreliable even for a Microsoft product. Ever since, I always cringe when emulation is proposed, always preferring native execution. (Incidentally, this real-world outcome proved to be a godsend for both Microsoft and Intel.)
However, this time with the FreeBSD compatibility layer it really is different. Linux system calls are so similar to FreeBSD system calls that Linux applications often run faster on identical hardware with FreeBSD under Linux emulation than they do natively under Linux. I know of at least one major commercial product in final development reliably emulating Linux executables under BSD Unix, with performance improvement as a surprise bonus. (Obviously the switch to BSD Unix was mandated because of licensing issues.)
With respect to the need to port BSD Unix to a broad range of hardware platforms, clearly it would be in WIND’s interest to encourage the FreeBSD community in this regard. On the other hand, not all of WIND’s CofE semiconductor partners may be willing to wait until their number is drawn. Not to worry. Since the CofE teams are sponsored by the respective semiconductor company, the choice ultimately depends on each semiconductor company’s resources and perceived opportunities with BSD Unix. (This is why the CofE concept is so powerful for mating software with hardware. It is based on a microcosm of the free capital market, which has proved superior to all alternative systems for allocating resources.)
The deeper I look, the better I like the BSD Unix acquisition, and the more I can appreciate the WIND’s bubbling enthusiasm during yesterday’s conference call.
Allen |
