Date: Wed, 27 Aug 1997 10:16:41 -0400 (EDT) From: Christopher Petrilli <petrilli@amber.org> To: Kyle Mestery <mestery@winternet.com> Cc: Peter Stubbs <peters@gil.com.au>, smp@FreeBSD.ORG Subject: Re: A how does it work question. Message-ID: <Pine.BSF.3.95q.970827101440.20292B-100000@chaos.amber.org> In-Reply-To: <Pine.GSO.3.96.970827091116.7668F-100000@tundra.winternet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
In the end, I would think that software would be the conjstriction point (in fact seperate memory makes it an MPP system, not an AMP/SMP system). This is the concept behind ccNUMA, etc... Because of the nature of the FreeBSD kernel (and I suppose the probably applies to Linux, but don't know), all I/O is threaded thru the #0 CPU, and thereby it becomes a HUGE bottleneck. Am I correct? This was what I was taught was the definition of a AMP system, was that a single CPU controlled all I/O on the system, which is why you have to go SMP to scale to X, and MPP to keep going from there. Chris On Wed, 27 Aug 1997, Kyle Mestery wrote: > > On Wed, 27 Aug 1997, Christopher Petrilli wrote: > > > Because of this, wouldn't it be appropriate to say that FreeBSD is an > > Assymetric MP, not Symmetric? Symmetric means that the kernel runs on > > each processor, and there is no "one processor" which controls exclusivity > > to the hardware. > > > The symmetric/assymetric refers to the hardware. This is from Curt > Schimmel's book on Cacheing and MP systems: > > "To be considered an SMP system, all of the CPUs in the system must be > connected to a single bus and share a common pool of memory and share > access to all I/O devices. Each CPU must also have equal access to the > memory and bus, and some form of bus arbitration must be considered. The > bus arbitration is usually entirely up to the hardware." > > In other words, an assymetric-MP system would be a system where each CPU > had it's own memory, own bus, etc. > > Kyle Mestery > StorageTek's Network Systems Group > 7600 Boone Ave. N., Brooklyn Park, MN 55428 > mesteka@anubis.network.com, mestery@winternet.com > > > > Chris > > > > On Wed, 27 Aug 1997, Kyle Mestery wrote: > > > > > On Wed, 27 Aug 1997, Peter Stubbs wrote: > > > > > > > I've been forced to sell my soul lately by doing a couple of MS Win > > > > NT courses ( mouths to feed etc.. ). It seems that NT runs a > > > > seperate instance of the kernel on each CPU present to provide it's > > > > SMP support. > > > > > > > > Is this the way FBSD smp does it? > > > > > > As far as I know, no. There is only one copy of the kernel running. At > > > present, access to the kernel is only allowed for one CPU (except for a > > > few areas), Steve has been working on making it reentrant. > > > > > > > Is this the only way to do it? > > > > > > No. Having the kernel be reentrant is another way. This requires the > > > correct lock "pushdown" into the kernel. Instead of one giant lock, > > > subsystems can each have their own lock, allowing multiple processors to > > > be in different sections of the kernel. This allows for increased > > > parallelism. > > > > > > > Doesn't this mean that lots more memory would be used keeping data > > > > for 2 kernels? > > > > > > > I would assume so, but I dont know for sure. > > > > > > Kyle Mestery > > > StorageTek's Network Systems Group > > > 7600 Boone Ave. N., Brooklyn Park, MN 55428 > > > mesteka@anubis.network.com, mestery@winternet.com > > > > > > > > > > > > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.95q.970827101440.20292B-100000>