| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1161.1 | A simplified explanation, it's more complicated than this | IJSAPL::OLTHOF | Henny Olthof @UTO 838-2021 | Wed Jun 10 1992 08:31 | 17 |
| Jeanne,
The 10% is no coincidence. Each user submits a transaction each 10 seconds
(on average). So if you have a system that does 10 TPS, you must have 100
users to generate these transactions. The 10 seconds 'think-time' is a
figure written down in the TPC-A specification.
The way to get to the number of TPS is get a system, load it with users and
get it running. Measure the responsetimes. Increase the number of users
(also the size of the database etc) until the point, that no longer
90% of all transactions finish within specified time. The last 'valid'
TPS rating within the sepcified time (don't recall wether that's 1 or 2
seconds) is the official one. So yes, 90% of all transactions finish within
the specified time.
Regards,
Henny Olthof
|
| 1161.2 | This isn't a DEC-specific thing | COOKIE::BERENSON | Lex mala, lex nulla | Wed Jun 10 1992 16:28 | 3 |
| Just to be clear, you will see this relationship in ALL TPC-A benchmarks
from ALL vendors. The TPC-A specification requires that it be run this
way.
|
| 1161.3 | Is the spec on-line? | VAOU02::JBLAIR | | Wed Jun 10 1992 19:47 | 4 |
| Thanks for the quick reply.... Is the TPC-A spec on-line? I want to
know more about this. Thanks a bunch,
jeanne
|
| 1161.4 | Check with our man in the committee | IJSAPL::OLTHOF | Henny Olthof @UTO 838-2021 | Thu Jun 11 1992 08:32 | 10 |
| Jeanne,
I suggest that you contact our representative in the TPC committee. His name
is Walt Kohler (@TAY). If you can get a hold of one of the published
performance reports, that will give you also some information. The description
of the benchmark is in the first chapters. We used to publish these things,
so check at your documentation center. These reports could also be ordered
through VXT LOS in the past, don't know if this still applies.
Henny Olthof, TP-DB The Netherlands
|
| 1161.5 | We have 3 reps to TPC | COOKIE::OAKEY | Picard/Riker '92 | Thu Jun 11 1992 16:08 | 14 |
| Jeanne,
We have 3 representatives with TPC
Walt Kohler (as -.1 said)
Amitabh Shah
Mike Brey
Don't expect an answer from any of them this week. The bi-monthly TPC
meeting is in Paris this week and they are all attending.
Another place to collect the actual various TPC specs is through the
Council directly, since they administer them. I was not able to
immediately track down TPC's address, sorry.
|
| 1161.6 | its in there. | WILBRY::NEEDLEMAN | | Fri Jun 12 1992 19:46 | 4 |
| look in tpsdoc::sys$public:[tp_performance] for much of your requested
data.
Barry
|
| 1161.7 | The Benchmark Handbook | KETJE::GERARD | Jean-Paul GERARD - EIS - Brussels | Wed Jun 24 1992 17:11 | 16 |
| Hi Jeanne,
You could also buy the benchmark bible:
The Benchmark Handbook
For database and transaction processing systems
Edited by Jim Gray
Morgan Kaufmann Publishers
2029 Campus Drive, suite 260
San Mateo, CA 94403
ISBN: 1-55860-159-7
Jean-Paul
|
| 1161.8 | Application perspective on .0 | TPSYS::SHAH | Amitabh Shah - Just say NO to decaf. | Wed Jun 24 1992 19:49 | 37 |
| Re. .0
> I have noticed that our TPC-A results have an interesting relationship
> between the tpsA-local value and the # of users. In each case the tpsA
> value is approximately 10% of the # of users. I wonder why?
Jeanne:
It helps to picture the application TPC-A is trying to model. Imagine
a bank having a number (say, B) branches. Each branch has *10* tellers,
each sitting in front of a terminal, and entertaining customers who
come in for transactions. Thus, for B branches, there are 10B tellers,
each corresponding to a user in the database.
Now, TPC-A has been designed so that each teller services 1 customer
approx. every 10 seconds; thus, each branch does approx. 1 transaction
every *1* second.
Thus: the tpsA number reported is approx. equal to the number of
branches supported by the database (ceiling(tpsA-number) = B), and the
number of users (tellers) is approx. 10 times this number.
Note that TPC-A does not disallow having more branches than the
reported tpsA number, but no one would want to do that from the price
/performance point (the amount of data storage required to be priced
is directly proportional to the number of branches).
TPC-A does not allow having fewer branches than the reported tpsA
number. Thus, with B branches, one can have a tpsA number in the
range (B-1,B] only.
Hope this helps.
-amitabh.
BTW, all TPC specifications are available on-line in PostScript and
MS Word format by (anonymous) FTP from dg.com, directory /ftp/tpc.
|