What databases exist that make use of clusters of machines running
parallel libraries (PVM?) or RPC contructs? Any?
J. S. Jensen
> What databases exist that make use of clusters of machines running
> parallel libraries (PVM?) or RPC contructs? Any?
Art S. Kagel
> > What databases exist that make use of clusters of machines running
> > parallel libraries (PVM?) or RPC contructs? Any?
> Both Oracle and Informix have Clustered/loosely coupled MP versions,
> but, do not use either PVM or RPC. Oracle relies on its global lock
I'll keep looking...
J. S. Jensen
> > > What databases exist that make use of clusters of machines running
> > > parallel libraries (PVM?) or RPC contructs? Any?
> > Both Oracle and Informix have Clustered/loosely coupled MP versions,
> > but, do not use either PVM or RPC. Oracle relies on its global lock
> The reason I ask is that I would like to do scalability tests on
> exceptionally complex SQL statements. Results would be relative, thus the
> db doesn't particularly matter (at least too much :-). If there were any
> dbs employing PVM or RPC, then I could truly test with a large number of
> I'll keep looking...
> J. S. Jensen
Press Release Follows:
NCR SENDS TPC-D THROUGH THE ROOF WITH A 3 TB TPC-D
On Friday, June 19, NCR published the first-ever TPC-D benchmark at 3
TB, breaking industry records for high volume decision support
Performed on a 32-node 5150 Teradata system, the benchmark has set the
standard at 3 TB with these characteristics:
-- 32-node 5150, with 4 x 200 MHz CPUs, 1 GB memory per node
-- Disk Storage Ratio = 3.69
-- 3 streams in the Throughput Test
-- Full Data protection with RAID5
Notable about this new TPC-D are the following:
-- NCR/Teradata once again achieved a feat unreachable by any other
platform today. Competitors who have managed to produce a 1 TB TPC-D
benchmark have done so only by totally exhausting themselves, even on
their "futures" platforms. This higher volume point is simply out of
reach for them.
-- This 3 TB benchmark configuration (32 nodes, 128 x 200 MHz CPUs, 32
GB memory) was substantially less hardware than used by IBM in their
recent 1 TB results (48 nodes, 192 x 332 MHz CPUs, 144 GB memory),
underscoring Teradata's ability to perform well with minimal hardware.
-- This very low disk storage ratio of 3.69 includes complete data
protection and sets to rest any speculation that Teradata solutions have
to carry high disk overhead, as do our competitors'.
-- The 3 concurrent query streams executed in 11% less time than the
single stream would have, had the single query stream been run 3 times
serially. This 11% scale-up substantiates the strong multi-user
capabilities that Teradata consistently delivers on large and on small
systems, at low or at high volumes, with few or with many users.
-- Single = 21710.6 sec
-- 3-stream = 58,000
-- Single times # of streams = 21710.6 * 3 = 65131.8
-- Multi-time / 3 serial single streams = 58,000 / 65131.8 = .89 or 89%
-- Scale-up = 100% - 89% = 11%
-- Comparing last year's 1 TB produced by NCR/Teradata, on a similarly
configured 32 node 5150 platform with this recent 3 TB, the QppD per
node performance has increased from 380 per node (12,149 / 32 = 380) up
to 451 per node (14,446 / 32 = 451). As the same level of parallelism
was active on each node, and the hardware was equivalent, this 20%
increase comes primarily from Teradata database and application
-- While other vendors cannot even take this giant step up, Teradata is
nowhere near exhaustion with this 3 TB benchmark. Both the hardware
platform (which can easily scale to 96 nodes and beyond, as we have seen
in production Teradata systems), and the Teradata database have been
proven very capable of taking on greater and larger challenges. Nor was
the system under any particular stress during this execution.
The good news for customers is the NCR/Teradata platform that produced
this 3 TB benchmark is available today, and was composed of current
hardware and software. Further, it demonstrates a low disk storage
ratio and the same simple 2-page database setup representative of
Teradata systems around the world.
Our research group has 3 RS/6000's (two model 350s and a 32H) that
I'd like to set up to allow us to do parallel computing, like one
can do on the SP/2 here at Cornell. Does anyone know if something
like this is possible ? If so, what software is required and
where do I get it ?