Re: revivesa status 2008/07/09

[Gary Thorpe <gathorpe79%yahoo.com@localhost>]

--- On Fri, 7/11/08, Mindaugas Rasiukevicius <rmind%NetBSD.org@localhost> wrote:

> From: Mindaugas Rasiukevicius <rmind%NetBSD.org@localhost>
> Subject: Re: revivesa status 2008/07/09
> To: tech-kern%netbsd.org@localhost
> Received: Friday, July 11, 2008, 12:18 PM
> Andrew Doran <andrew%hairylemon.org@localhost> wrote:
> > It has to be said, I have not seen a convincing
> explanation as to why this
> > is desirable or in the best interests of NetBSD as a
> product.
> > 
> > SA threading in NetBSD has serious problems and
> drawbacks. For example:
> > 
> > - it works only on a handful of architectures, eg x86.
> > - in most tests its performance is demonstrably
> inferior to 1:1.
> > - it's completely unreliable, even opening the
> machine to DOS attacks.
> > - it has architectural, code quality and code
> maintenance issues.
> > - it completely lacks any kind of real-time support.
> > 
> > In its current form SA threading is a regressive
> proposition. Even if all
> > the remaining issues are addressed, what benefits
> would it bring over and
> > above 1:1 threading?
> 
> With all respect to Bill's work and his good ideas - he
> really improved SA
> lot, and put a lot of effort on this - I fully agree with
> Andrew. Besides:
> 
>  - SA increases code complexity a lot (in our threading
> subsystem);
>  - does not support thread affinity features;
>  - have not seen that performance of SA is better than 1:1
> even on very
>    CPU-bound workload;

The advantages of SA are supposed to be on *I/O* bound workloads because it can 
reduce overhead due to kernel context switches by doing userland switches where 
appropriate (according to previous benchmarks, research etc.), but I could be 
wron on this.

> Also, I believe SA might be potential security problem
> (over-complicated code
> tends to be breakable - consider a lot of threads with
> continual blocking).
> That is why I suggested to make SA at least a kernel
> option. My suggestion to
> make SA a kernel option was exactly because of that. This
> also would reduce
> the modifications to the existing threading subsystem.
> 
> -- 
> Best regards,
> Mindaugas
> www.NetBSD.org


      __________________________________________________________________
Yahoo! Canada Toolbar: Search from anywhere on the web, and bookmark your 
favourite sites. Download it now at
http://ca.toolbar.yahoo.com.