[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: SMP



>If you look on bsdtoday.com there is an 'article' about SMP on NetBSD
>(which is really just a link to their mailing list archive).  See
>"http://bsdtoday.com/2000/August/News266.html"; for more info.

I was a little concerned about the way they were attacking the SMP issue.
While I understand that the article was a very terse report I think they
have a long road ahead of them in getting the locks working.

If you were writting a system from scratch it would be fairly simple (not
easy) to include the locking during the design phase. Now the code must be
examined for each race condition and any contention for resources. Then
careful coding must be done to prevent deadlocks. 

Linux avoided some of the problems with this kind of refit by defining a
single lock that covered a large number of resources. You don't have a
problem with deadlocks when there is only one lock. The problem with this
design is that you kill performance with lock contention. They have since
added additonal locks to provide more granularity over resources.