home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Power Tools 1993 November - Disc 2
/
Power Tools Plus (Disc 2 of 2)(November 1993)(HP).iso
/
hotlines
/
gsyhl
/
seq1002a.txt
< prev
next >
Wrap
Text File
|
1993-09-20
|
4KB
|
68 lines
HP Internal Use Only
In June 1993, Sequent announced 1002.37 tpsA at $9,313/tpsA on a 2-system
cluster with a total of 46 processors. They achieved this result using
Oracle7 Parallel Server (OPS).
While this number is the highest TPC-A number published, there are
several relevant points to remember and use in competitive situations.
- Bottom line - Sequent could not match our record-setting 710 tpsA
with one system and had to resort to a 2-system cluster to beat our
performance. This proves that Sequent has lost the performance game to
HP! This result, while sounding impressive, does not achieve anything
new with respect to performance (it does show that they can run Oracle
Parallel Server). Most any vendor can, by using a transaction monitor
that supports two-phase commit, produce results at this level through
multiple clustered systems. Our labs have already done this
internally. HP can exceed Sequent's 46-processor result by clustering
only three H70s or two 890s. They achieved about 22 tps per processor
while we can achieve over 150 tps per processor in a similar cluster.
- Sequent's cluster is a niche solution. This solution will only
appeal to the minority of customers who need a cluster to fill specific
needs like high-availability. Clusters are not generally used today
for real performance growth for real-world applications.
- Sequent's price/performance is uncompetitive with HP. At
$9,313/tpsA, they are worse than our 890/4 price/performance. With
the new system pricing, the 890/4 is at $6,767/tpsA. Their solution
has 38% worse price/performance and has a total cost of over $9.3
million! The CPUs alone list for $2,520,000 versus $360,000 for the
890/4.
- We have achieved better performance on one system. Our labs have
already beaten Sequent's number on an 8-way 890 in internal testing.
This does not even account for future chip enhancements over the
current Emerald chips.
- Sequent will likely be unable to provide references for such a large
configuration. Have your customer challenge Sequent to provide real
customers using 46 processors in a 2-system Sequent cluster. They may
even have trouble providing references for a 1-system solution with 24
processors.
- HP has a better in-cabinet growth path. To approach this
performance-level, Sequent customers must add a second box while 890
customers can do board upgrades. When Sequent customers try to move to
Pentium, they will be required to recompile all their applications to
achieve more than a 20% performance improvement.
- TPC-A was not designed to measure real application performance.
TPC-C is a superior benchmark (more complex and realistic). HP has
TPC-C results and Sequent doesn't. Where are Sequent's TPC-C results?
In clustered environments, TPC-A is even less useful. Performance in
clustered solutions typically bottlenecks in the network connection
between the clustered systems. The more transactions (and the greater
their complexity) that cross this network, the more performance
degrades. To meet TPC-A specifications, only about 7.5% of Sequent's
transactions crossed the network to the other system. Since real
applications typically have much more than 7.5% remote transactions (up
to 50% and more), real application performance with a Sequent cluster
could degrade severely! In addition, the TPC-A transaction is
extremely simple. Real-world transactions are much more complex and
will cause performance degradation over this network link even more
quickly than TPC-A would show.
***********************************************************************