| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1120.1 | | KONING::KONING | Paul Koning, B-16504 | Thu Oct 21 1993 11:19 | 5 |
| There is no such thing. SpecInt measures application compute performance.
It doesn't apply to networking performance. Your customer has been given
nonsensical information by the other vendor.
paul
|
| 1120.2 | CPU utilization | TAV02::KATZAV | | Thu Oct 21 1993 16:32 | 8 |
|
Maybe the other vendor took the CPU consumption and translated it
to the spec as a percent of the total spec?
What are the cpu utilization on DEC3000-600 with FDDI to transfer
1Mbyte/sec ??
Thanks,
Shimon.
|
| 1120.3 | | NETRIX::thomas | The Code Warrior | Thu Oct 21 1993 16:55 | 6 |
| Using what protocol? From the file system or generated data? Aligned
or unaligned data? To what type of system?
Aside from that, since (at least under OSF/1) a 3000/500 can almost
saturate a FDDI ring with about 30-40% CPU left over, I would expect
a that 3000/600 should be able to even better than that.
|
| 1120.4 | | KONING::KONING | Paul Koning, B-16504 | Fri Oct 22 1993 13:21 | 12 |
| Re .2: even so, that makes no sense. All that's being displayed is a great
ignorance of benchmarking and performance measurement.
"specmark" is a number obtained from executing a very specific set of tests
in a very specific environment. It CANNOT be applied to any other test,
especially not to tests of a very different kind, such as network performance.
It would make as much (or rather, as little) sense to ask "what is the capacity
of your system disk, expressed in specmarks", or "how big is your memory,
expressed in specmarks". (Or "how far is it to Cleveland, expressed in pounds")
paul
|
| 1120.5 | to explain Paul's response | MUDDY::WATERS | | Fri Oct 22 1993 13:53 | 19 |
| > Maybe the other vendor took the CPU consumption and translated it
> to the spec as a percent of the total spec?
Yes, you can translate percentage CPU utilization for 1 MB/s FDDI (using
specific protcols/applications) into SPECint92 *of a specific computer
model*. What you've done is to take a fairly clear performance metric,
percentage CPU utilization for 1 MB/s of xxx transfer over FDDI
on computer model yyy [where xxx is the network application]
and turn it into a less clear metric
SPECint92 CPU utilization for 1 MB/s of xxx transfer over FDDI
on computer model yyy whose peak SPECint92 was zzz on date dd-mmm-yy
The new metric takes more lines to explain than the old one; it relies
on more variables (both network software version *and* SPEC benchmark
version and compiler version). The new metric should be avoided; the
old percentage-CPU metric has more direct meaning.
|
| 1120.6 | FDDI white paper | TAV02::KATZAV | | Fri Oct 22 1993 14:16 | 6 |
|
Got the message.
Any white paper that has performance numbers in it ??
-Shimon.
|
| 1120.7 | FDDI NIC White Paper in note 1070 | MUDDY::WATERS | | Fri Oct 22 1993 14:52 | 1 |
| Is note 1070 any help?
|
| 1120.8 | | KONING::KONING | Paul Koning, B-16504 | Fri Oct 22 1993 18:52 | 11 |
| Re .5: I disagree, since SpecInt is a composite of several test programs,
and the ratio of the running speeds of program x doesn't necessarily match
the specint ratio. (This is because of the way the system architecture
affects a particular program, the way the compiler optimizes that particular
program, etc.)
Therefore the specint number doesn't predict the performance of any of
the programs in the test suite. It DEFINITELY does not predict the performance
of any other program.
paul
|
| 1120.9 | | MUDDY::WATERS | | Sun Oct 24 1993 12:09 | 7 |
| >Therefore the specint number doesn't predict the performance of any of
>the programs in the test suite.
No, you don't disagree with me. We're both saying that SPECint92
predicts nothing; it only measures itself (a benchmark suite). I did
not claim that SPECint92 does anything other than measuring its own
performance on one machine.
|