From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from parasite.irisa.fr (parasite.irisa.fr [131.254.12.47]) by kvack.org (8.8.7/8.8.7) with ESMTP id DAA09272 for ; Tue, 9 Dec 1997 03:02:30 -0500 Subject: Re: VM ideas (was: Re: TTY changes to 2.1.65) References: Mime-Version: 1.0 (generated by tm-edit 7.95) Content-Type: text/plain; charset=US-ASCII From: David Mentre Date: 09 Dec 1997 08:56:31 +0100 In-Reply-To: Rik van Riel's message of "Mon, 8 Dec 1997 13:23:30 +0100 (MET)" Message-ID: Sender: owner-linux-mm@kvack.org To: H.H.vanRiel@fys.ruu.nl Cc: Joerg Rade , linux-mm List-ID: Hello Rik, Rik van Riel writes: > On Thu, 4 Dec 1997, Joerg Rade wrote: [snip] > > My own vote for "most worthwhile hardware for memory management" > > is sub-page protections in the TLB, e.g., 1-KB independently-protectable > > units withing 4KB or 8KB pages, like the ARM 6xx series. > > Swapping is mainly slowed down by _latency_ problems. The [...] The interesting point of Joerg is that he see the TLB mecanism as a more general mecanism than just to solve swapping problems. > > These features are desirable for a lot of things, including: [snip] > > 3. distributed virtual memory, > DIPC is out... check your favorite web site... [snip] As I'm a little involved in Distributed Shared Memories (with a PhD ;), I couldn't let such an opportunity happen without talking. I totally agree with Joerg. One problems with DSM is that you must track user memory accesses to maintain coherency. Unfortunatly, fine grain access like cache line is not available to the average system programmer. Therefore sub-page protection could be very useful. Regarding DIPC, I think we could improved a little the coherency protocol. One big advantage of DIPC is that it provide code, and you can't lie with code. :) I hope I'll have more code in the future to explain my point of vue in DSM. david -- David.Mentre@irisa.fr -- CAPS research team - Irisa - France Personal info: http://www.irisa.fr/prive/mentre/ Professional info: http://www.irisa.fr/caps/PEOPLE/David/