Slashdot Mirror
Solaris 10 to be Open Source
An anonymous reader writes "It looks as though Sun is going to open source their new Solaris 10 operating system. It seems to include eveything except some device drivers. They plan to model the Darwin and Fedora projects. Sounds very interesting."
What does SUN do anymore? If they're open sourcing Solaris, obviously they're looking to get the community involved in developing it. They're also starting to ship some x86 servers (Opteron and Xeon), so are we eventually going to lose the Sparc processors as well? What does that leave Sun with? Java?
I wonder how they'll handle the Unix(TM) code in there and all the various other contributed stuff from Samsung etc.
I guess it's easier if they forget about CDE/X11 etc but it will be interesting to see what open source licence they use and how they handle 'other peoples' code in SOlaris 10.
Of course they could have removed all the Sys V R5.4 code, but without doing this unsing clean room conditions SCO could have a wondrful time in court.
Just wondering??????
Schwartz invoked the precedent set by Sun's popular Java programming language. [...] We need to now take the model with Java and bring it to Solaris.
A problem that Schwartz wants to avoid is having Solaris splintered into different distributions like Linux, which he said creates application incompatibilities. Going the way of Linux-type licensing, he suggested, creates open source but not open standards.
Watch great movie opening scenes!
Let me explain Oracle's conversion to Linux.
You are right that until recently the "reference platform" for large Oracle installations was Solaris, and Oracle would run efficiently and scalably across tens of processors.
Then Oracle invented parallel databases. Their first attempt, Oracle Parallel Server in 8i was horrible, held together with string and bubble gum. Nobody used it.
Then they came out with the next version, 9i RAC, which was quite a lot better. But any attempt to run a read/write database across a number of servers is always going to be limitted by the speed of the interconnects, so it is still far preferable to run 9i non-RAC on a large server than RAC across multiple machines. So enter Oracle's love affair with Linux.
Oracle have taken to pitching 9i RAC solutions on Linux as being the "cheaper" alternative to running on a big Solaris box. The rational is simple: the customer either pays Oracle for 9i non-RAC and Sun for a big box, or they pay Oracle for 9i RAC and implement it on commodity x86 hardware running GNU/Linux - obviously they prefer the second solution because they get more money from a similar sized cake.
The snag is that 9i RAC doesn't scale well, because of the previously mentioned interconnect latency issue. They will quote you impressive figures which are the result of:
a) picking benchmark examples which are naturally going to scale well across boxes - where the data sets are already well partitioned
b) comparing RAC on two nodes to a single node running RAC - the true comparison would be with a single node running 9i non-RAC (which is loads faster).
So don't imagine that this is Oracle having been converted by any sort of technical merits - they are being driven simply by ways of maximising their revenue stream.
Depends a lot on what you're doing. SPARC might be OK at high-throughput jobs, but IA32 and PowerPC just smash it to little bits for things that are less sequential.
Also, Solaris' local filesystem (UFS) gets the pants beat off it by EXT3 (and, to a lesser extent, AIX JFS2). Even if you turn on journalling, which makes for a nice speed boost on Solaris 8 and up.
In fact, for local file I/O, you're better with Solaris on IA32 than Solaris on SPARC.
I'm not actually sure what SPARC hardware is good for these days. Every time I benchmark something, it loses. Granted, our best SPARC machine is an 8-way UltraSPARC-III 1.2 GHz. So maybe a faster SPARC chip might keep up with PowerPC and Intel a little better.