OpenSSI 1.9.6 in the making

Roger Tsang <[email protected]> Tue, 10 Nov 2009 08:05:45 +0000
Newsgroups gmane.linux.cluster.ssic.devel
Message-ID <[email protected]>
Hi,

These are just some thoughts I'd like to share.

As you know the original plan was to release 2.0.0pre3 as 1.9.6 around
first half of year 2009.  Early in the year we decided to rename the
release to 1.9.6 due to the amount of changes to the core.  However
one criteria I had in mind for making this release was improved
stability from the previous release.  At the time I did not feel we've
met this criteria.  At the same time some people felt that porting to
2.6.18 is more urgent, but I strongly believe the core should be
stabilized first to ensure we won't have to deal with outstanding core
bugs that could impact our porting effort.

So we ended up focusing our effort fixing core bugs.  At the same time
I took the opportunity to increase the performance of OpenSSI.  Thanks
to John Hughes we also identified some useful regression tests.  Today
most of the bugs in our tracker are closed or fixed.  We also
completed more than a few OpenSSI enhancements / feature requests in
our tracker.

During the course of the year we've seen almost 20,000 lines of code
(LoC) changes at the core.  The net change is less but this is still
5-10x more than the previous year.  I haven't really looked up the
numbers; that's my impression.  Worthy of mention in addition to
system latency performance improvements is the time and effort taken
to address the more obscure bugs only seen on multi-core / SMP
machines and preemptive kernels.

Though rarely mentioned I believe one of the major technical obstacles
to reaching stable was OpenSSI-1.9.x had been suffering the same class
of problems the openMosix project (now defunct) had long struggled
with and that is SMP bugs.  If I'm not wrong the developers who have
forked openMosix are still dealing with these problems.  So I am
thankful that most of our OpenSSI developers and designers kept this
in mind when porting to Linux and am grateful that we are able to
overcome the remaining issues considering our OpenSSI code base is
much larger than openMosix and considering we've integrated some
openMosix code (eek!).


Regards,
Roger

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july