|
|
| Well, He Should Know. |
 |
|
Friday, June 20 2003 @ 07:27 PM EDT
|
For analysis of the Byte interview with Chris Sontag of a few days ago by
Greg Lehey,who is a FreeBSD guy and who worked in IBM's Linux Technology Center, writing "a clone of the AIX Journalled File System, the predecessor of the JFS ported by the JFS for Linux project," go here:Specifically, Sontag believes the "SCO technologies" which were misappropriated into AIX, IRIX, and the derivative UNIX-alikes (including Linux) are:
JFS ( Journalling File System). -- This is unbelievable. Did Sontag really say that? To this day SCO doesn't have code of that name. JFS was developed by IBM. To judge by the header files, it appears to base heavily on the Berkeley (BSD) Fast File System. NUMA (Non Uniform Memory Access), a SGI/Stanford collaboration. -- They don't even claim to have anything to do with this one. That's a good thing, too: NUMA is very much a hardware issue. SCO doesn't do hardware. RCU (Read-Copy-Update ). -- This is an algorithm. There's nothing in here which required, or could even benefit, access to the source code of an implementation in a differently structured system. SMP (Symmetrical Multi-Processing). -- SMP predates UNIX. SCO was a latecomer in implementing SMP, and as far as I can tell none of its license holders use the SCO version: they all have their own implementations, all of which are superior to the SCO implementation. The link shows Linux's implementation, which was initially very much a primitive "Giant Kernel Lock" implementation that has been around for decades, Later versions of the Linux SMP implementation differ strongly from any UNIX implementation. And on his blog, he says IBM carefully kept AIX and Linux coders separate: Having worked on Linux for IBM, I can state categorically that the separation between AIX and Linux is complete. Nearly all of the people working on the Linux kernel have no access to AIX source code. It's theoretically possible that some people do have such access, though I know of nobody, but IBM has guidelines for that case, just to be on the safe side: don't read AIX code and write Linux code in the same place. Read the code, go elsewhere and write. Even this, though, would hardly be useful: AIX is UNIX, Linux is Linux. The kernels have such completely different structures that any code import would be a waste of time: it's easier to write it from scratch.
SCO's talk about SMP scalability is nonsense. They don't have any useful technology in that area. Linux scaled better than UNIX System V long before IBM came on the scene. It's true that IBM is doing a lot of work in that area, but it's based on the PowerPC architecture, and SCO's version of System V doesn't support that platform. So what use would the code be, even if it were a drop-in replacement on the i386 platform?
|
|
|
|
|
