SUMMARY Speed of Sparcserver 1000
Sorry for being late with this summary, but I was still trying to
collect the infos. The original mail was :
> Hi,
> really disappointed about the performance of our new SPARCserver 1000
> for a single user, could give someone me ideas about the relation
> between SPECint92 and SPECrate_int92. Have someone out there experience
> how to speed up the machine acting more as a single-user workstation
> as a server. Any idea could be helpful.
System Configuration: Sun Microsystems sun4d
Memory size: 128 Megabytes
2 CPU's
4*500MB
Just not recognizing the little 'rate' between SPEC and int92 in
the performance guide of sun, I compare the SPECint92 of my little SPARC10-20
,which is 39.8, with the SPECrate_int92 of the SPARCserver1000, which
is 2730 with 2 CPU's. So it seems logical to divide this with 2 (CPU's).
So the SPARCserver performance seems to be more than 30 times better, than
the little SPARC10. ( Be sure, I havn't believe to see that really ;-)
For the performance Test I use the 'BYTE UNIX Benchmarks Version 3.11'
Here is what I get for one Test.
SGI Indigo2 with IP22 5.6
SGI Indigo with IP20 4.5
SPARCserver 1000 3.1
E&S ESV3 2.6
SPARC10-20 2.1
NeXTStation TurboColor 1.4
The Baseline is a 386/33
That's why I was dissapointed. But looking to the little 'rate' it's
going clear now more and more.
The SPARCserver 1000 carries 50Hz SuperSparc processors. So the performance
for a single task is identical with the performance of a SPARCstation10-51,
until you use multiple threads. The SPARCserver 1000 show his power if you
use it really like a server should used. The next step was a test how much
processes could work at the same time. I try how often I could start the
benchmark in the background at the same time. The SPARCstation10-20 smiles
with no more processes after 3 test working. The Indigo give up after 5.
A SGI 380VGX (i.e. 8*R3000) crash after 8. I stop the test on the SPARCserver
1000 after 20 benchmarks works well together.
To get the feeling into numbers I run 5 benchmarks at the same time on
the server and on a SGI Indigo with IP20. Here are the average results
of the benchmarks:
SPARCserver 1000 1.7
SGI Indigo with IP20 1.0
BTW, the 20 benchmarks on the server results in an average of 1.0
Conclusing:
The SPARCserver 1000 is not the hotest thing you could get for a single
task, but if you have a lot of jobs to do at the same time, he will be a
good-naturedness machine with enough power to do all jobs well.
Every remark to this Summary will be welcomed. Also if you have some
more question about my (till yet short) experience with the server don't
hesitate to mail me.
Thanx to everybody who helped me with his knowledge about the SPARCserver1000.
especially to:
jaa101@deakin.anu.edu.au (James Ashton)
blymn@mulga.awadi.com.au (Brett Lymn)
griffin@lehman.com (al griffin)
Dirk.
--------------+----------------------------------------------------------
| | |_|_|_|_| Fraunhofer Institute for Computer Graphics |
|__|__|_|_|_|_| Department Animation & HD Image Communication |
| | |_|_|_|_| |
|__|__|_|_|_|_| Dirk Boenning email : boenning@igd.fhg.de |
| __ | __| Wilhelminenstr. 7 phone : (+49) 6151-155-141 |
| /_ |/_ /_ | D-64283 Darmstadt fax : (+49) 6151-155-199 |
|/ / / /_/ | Germany |
|-------------+---------------------------------------------------------|
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:16 CDT