Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 13 Jun 2006 09:11:10 -0400
From:      John Baldwin <jhb@freebsd.org>
To:        freebsd-smp@freebsd.org
Cc:        "Chris H." <fbsd@1command.com>
Subject:   Re: Concluding the SMPng project
Message-ID:  <200606130911.10775.jhb@freebsd.org>
In-Reply-To: <20060612141235.3f7woozpc4888ksc@webmail.1command.com>
References:  <20060606195938.GA6581@xor.obsecurity.org> <20060612141235.3f7woozpc4888ksc@webmail.1command.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Monday 12 June 2006 17:12, Chris H. wrote:
> Quoting Kris Kennaway <kris@obsecurity.org>:
>=20
> > Several of us have been discussing recently whether it is time to
> > bring the SMPng project to a formal conclusion.
> >
> > According to the SMPng project webpage,
> >
> >  "The end goal of the SMPng Project is to decompose the Giant lock
> >  into a number of smaller locks, resulting in reduced contention (and
> >  improved SMP performance)."
> >
> > Thanks to the hard work of many developers over the past ~6 years, this
> > goal is now complete.
> >
> > While Giant has not been completely eliminated from the kernel and
> > several subsystems are still giant-locked (notably ipv6, tty, and CAM,
> > although work is in progress on all of these fronts), kernel profiling
> > traces show that for many real-world application loads the Giant lock
> > is simply no longer a factor in the performance of the SMP kernel.
> >
> > See e.g.
> >
> >  http://www.bsdcan.org/2006/papers/FilesystemPerformance.pdf
> >
> > for one such measurement of the extent of Giant locking in FreeBSD
> > 6.x; other real-world application workloads are similar.
> >
> > Some of the benefits of formally concluding the SMPng project are:
> >
> > * The focus of SMP development work has largely changed from "break up
> > Giant everywhere" to "carefully measure the effects of the locking
> > decisions that were made, and optimize for greater performance and
> > scalability".  This is a major milestone and should be announced to
> > the world, perhaps under the banner of a new "FreeBSD Scalability
> > Project".
> >
> > * For example, a number of us are looking very closely at the nascent
> > FreeBSD port to the Sun Ultrasparc T1, which provides 32 virtual CPUs
> > (4 threads on 8 CPU cores) on a single chip.  Optimizing for the new
> > generation of SMP hardware is going to be a major effort over the
> > coming year.
>=20
> Ahh, so the contributions made by the PIII & PIV CPU's were merely to
> obtain access to the Sparc systems, and the PIII & PIV will be relegated
> to the ubiquitous I386 scrap heap, as the future and ultimate goal of
> FreeBSD is to be Sun Microsystems. Pitty, FreeBSD has always provided
> such wide scalability. So easy to implement on so many architectures.
> I wish I had known it's agenda years ago. As I would not have spent
> so many years and so many dollars building *BSD based infrastructures.
> Perhaps I've misunderstood this announcement. But if not;
> good riddance.

Umm, no.  However, with multi-core becoming more fashionable, x86 systems
are going to start having more and more CPUs as well, so FreeBSD needs to
work on scaling up to more than say 4 CPUs.

=2D-=20
John Baldwin <jhb@FreeBSD.org> =A0<>< =A0http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve" =A0=3D =A0http://www.FreeBSD.org



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200606130911.10775.jhb>