This is Pyramids-Multiprocessor-based-on-MIPS-R3000 in view mode; [Up]
Date: Sun 27-Aug-1991 06:37:07 From: sfr@pdact.pd.necisa.oz.au (Stephen F. Rothwell) Subject: Pyramid's Multiprocessor based on MIPS R3000 I originally posted this to comp.arch and comp.sys.pyramid and someone suggested that I try comp.sys.sgi and comp.sys.mips, so here it is: A colleague of mine has heard that Pyramid have built a multiprocessor system based on the MIPS R3000 chip. He was wondering if anyone could tell him anything about the architecture of this machine. He is interested because he has been told that this is hard to do. The following is a quote from his letter to me: > > The following are some reasons that the MIPS R3000 is not a suitable > chip to build a MP machine: > > 1. cache consistency mechanism > A hardware mechanism,like hardware-cache-snooping is required > to guarantee the cache consistency among the caches installed > on each R3000. > R3000 and its family chips, however, does not provide any function > to support it. > > 2. bus lock/unlock mechanism > There is no instruction which performs atomic read-modify-write > sequences in multiprocessor and multithread systems. > For example,the i860XP has the lock/unlock instructions > to realize the same atomic primitives. > > 3. interrupt handling mechanism > There is no control mechanism to decide which CPU must be > interrupted by peripheral controlers. > > 4. cross-processor interrupts mechanism > This is required to realize the TLB and cache flush function, > clock synchronization, debugger interrupt and so on. > > There may be more reasons but these are the main ones. > > Of course, you can still build a MP machine using the R3000 but you have > to implement a lot of workarounds and degrade the performance of the system. Is there anyone out there who would like to comment on the above? Please send your comments to the news, as he can read news - he just can't post. Any mail sent ot me will be forwarded to him. Stephen Rothwell sfr@pdact.pd.necisa.oz.au NEC Information Systems Australia Software Development Centre, Canberra ====================================================================== Todoloo! -Pascal
These are the contents of the former NiCE NeXT User Group NeXTSTEP/OpenStep software archive, currently hosted by Marcel Waldvogel and Netfuture.ch.