siraj@endor.harvard.edu (Raad Siraj) (04/05/88)
Does anyone know of relational databases that take advantage of parallel machines? Last month Sequent Computer Systems gave a seminar on this topic in Boston, but there was no reference to any published papers. Moreover, the only printed news on this topic - that I found - was in the Government Computer News (Oct. 23 1987) with the title "Oracle hints at advances in parallelism". Ra'ad
daveb@llama.rtech.UUCP (Crack? No thanks, I've got a new CD player) (04/07/88)
In article <4317@husc6.harvard.edu> siraj@endor.UUCP (Raad Siraj) writes: >Does anyone know of relational databases that take advantage of parallel >machines? > >Last month Sequent Computer Systems gave a seminar on this topic in Boston, >but there was no reference to any published papers. Moreover, the only >printed news on this topic - that I found - was in the Government Computer >News (Oct. 23 1987) with the title "Oracle hints at advances in >parallelism". No commercial system I am aware of uses the parallelism possible in multiple-cpu systems to help a single user query very much. Some marginal improvement might be seen when an application program can run at the same time as a database server/backend, but since they would usually operate synchronously, it wouldn't really matter. There is less interest in helping the single user case through parallelism, as multi-user throughput is the major concern. Multiple CPUs can help immensely on multi-user system throughput, if the DBMS system is architected to take advantage of it. Specifically, systems that can run multiple DBMS processes against the same database will win. Systems with single/monolithic servers will lose, because they can't split processing of multiple concurrent queries across the extra cpus. INGRES through version 5.0 uses a DBMS process per user, and this works very well on multi-cpu systems. INGRES 6.0 is a multi-server architecture, where a server supports multiple users. This lets single or multiple servers run against a database. This wins on both single and multiple cpu systems because you can tune the number of servers to you application/harware mix. The two degenerate cases are one server per user as in pre-6.0 INGRES, or one server per machine or database as in some other systems. -dB "Do you programmers do sh*t like this a lot?" "Every g*d*mn day." {amdahl, cpsc6a, mtxinu, ptsfa, sun, hoptoad}!rtech!daveb daveb@rtech.uucp
markd@rtech.UUCP (Mark P. Diamond) (04/07/88)
From article <4317@husc6.harvard.edu>, by siraj@endor.harvard.edu (Raad Siraj): > Does anyone know of relational databases that take advantage of parallel > machines? > Ingres runs very effectively on the Sequent Balance and especially the Sequent Symmetry. Ingres's fastest performaces have been accomplished on the Balance (see the March 21 issue of FT Systems Newsletter lists some results of the TP1 benchmarks), and the just released Symmetry port of Ingres (Symmetry 4x over Balance) will be much faster. Mark <> <> <> <> Mark P. Diamond {sun, cbosgd, amdahl, mtxinu}!rtech!markd from Sequent Computer Systems onsite at Relational Technology
pavlov@hscfvax.harvard.edu (G.Pavlov) (04/10/88)
> In article <4317@husc6.harvard.edu> siraj@endor.UUCP (Raad Siraj) writes: > Does anyone know of relational databases that take advantage of parallel > machines? > .......... Moreover, the only > printed news on this topic - that I found - was in the Government Computer > News (Oct. 23 1987) with the title "Oracle hints at advances in > parallelism". > The Big O marketing machine rolls on.... greg pavlov, fstrf, amherst, ny