best performances for the java 6 JRE: linux or solaris?

[dao.hodac]

hello,

I have to deploy a java/CORBA server on a new host. I have to choose
between 2 configurations:

Conf#1
SUN, V240 2x(ultraSparc3@1,5GHz), 64Bits, RISK
Solaris 9

Conf#2
PC 2x(Opteron dualcore)@2.8GHz 64bits, SISK
Linux RedHat 4


which is the best for my java application, in terms of performances
and reliability? Is there some deployment tips about RAM, hard
drive... things to avoid... I must take into account ?

Java application:
no GUI
highly threaded
java 6 (I hope, 5 else...)
no developpement on this host

thanks.

 

[Lew]

hello,

I have to deploy a java/CORBA server on a new host. I have to choose
between 2 configurations:

Conf#1
SUN, V240 2x(ultraSparc3@1,5GHz), 64Bits, RISK
Solaris 9

Conf#2
PC 2x(Opteron dualcore)@2.8GHz 64bits, SISK
Linux RedHat 4


which is the best for my java application, in terms of performances
and reliability? Is there some deployment tips about RAM, hard
drive... things to avoid... I must take into account ?

Java application:
no GUI
highly threaded
java 6 (I hope, 5 else...)
no developpement on this host

I doubt there is a way to generalize /a priori/. Run sample apps that have
similar performance characteristics to your intended application(s), and
measure the results.

How important is performance, and how much difference must there be to matter?

Have you considered O.S. alternatives that may speed up performance, e.g., QNX?

One tip is to provide maximum RAM and very effing fast hard drives.

OTOH, aren't reliability and up-time ultimately more important than raw speed?
How durable are these machines? How likely for a RAM chip to fail? How
expensive is support for each platform?

If raw speed is so important, why limit yourself to those two platforms?

Are you using RAID drive configurations?

By the way, the acronyms are "RISC" and "CISC", not "RISK" and "SISK". They
are acronyms, so the spelling matters. GIYF.

 

[dao]

I doubt there is a way to generalize /a priori/. Run sample apps that have
similar performance characteristics to your intended application(s), and
measure the results.

How important is performance, and how much difference must there be to matter?

Have you considered O.S. alternatives that may speed up performance, e.g., QNX?

One tip is to provide maximum RAM and very effing fast hard drives.

OTOH, aren't reliability and up-time ultimately more important than raw speed?
How durable are these machines? How likely for a RAM chip to fail? How
expensive is support for each platform?

If raw speed is so important, why limit yourself to those two platforms?

Are you using RAID drive configurations?

By the way, the acronyms are "RISC" and "CISC", not "RISK" and "SISK". They
are acronyms, so the spelling matters. GIYF.

I am a software manager, my server has ran on a V240/solaris 5.9 for 2
years. Today, IT proposed me either to buy a clone machine or go to a
opteron/linux environment. I have to choose. The application uses the
CPU mainly for XML processing, and it is the bottleneck of it.

I don't know a lot about devices, butI have seen that the opteron is
almost 2 times faster, it is 4 processors (2 dual core), so I suppose
it is much faster (4 times?) than the V240.
Further more, the IT insists for the opteron because it is much
cheaper (2 times!).

I just want to check that is is a good environment (no surprise,
please!!!)

thanks for your answer (s).

 

[dao]

I doubt there is a way to generalize /a priori/. Run sample apps that have
similar performance characteristics to your intended application(s), and
measure the results.

How important is performance, and how much difference must there be to matter?

Have you considered O.S. alternatives that may speed up performance, e.g., QNX?

One tip is to provide maximum RAM and very effing fast hard drives.

OTOH, aren't reliability and up-time ultimately more important than raw speed?
How durable are these machines? How likely for a RAM chip to fail? How
expensive is support for each platform?

If raw speed is so important, why limit yourself to those two platforms?

Are you using RAID drive configurations?

By the way, the acronyms are "RISC" and "CISC", not "RISK" and "SISK". They
are acronyms, so the spelling matters. GIYF.

hi,

Sorry for the approximations about the description of the devices, but
i am not the IT.
Today, my application runs well on a SOLARIS 5.9 V240, and the IT
proposed me to either buy a new V240 or to get a linux/opteron.

I need to use the CPU for XML processing, that is the bottleneck of my
application (I have seen the linux solution is 2.8GHz, 2x2 procs...
therefore I though it was really faster).

Before sending my answer, I want to know this group point of view

PROS:
faster => I need CPU to highly process XML
cheaper => IT said it is 2 times cheaper

CONS:
migration

 

[Lew]

Today, my application runs well on a SOLARIS 5.9 V240, and the IT
proposed me to either buy a new V240 or to get a linux/opteron.

I need to use the CPU for XML processing, that is the bottleneck of my
application (I have seen the linux solution is 2.8GHz, 2x2 procs...
therefore I though it was really faster).

Before sending my answer, I want to know this group point of view

PROS:
faster => I need CPU to highly process XML
cheaper => IT said it is 2 times cheaper

CONS:
migration

Migration of Java code should be painless.

The application uses the
CPU mainly for XML processing, and it is the bottleneck of it.

The application uses a lot more than the CPU. The CPU is likely not the
bottleneck at all.

XML processing is subject to inefficiency in the software. Depending on what
software you use, who wrote it and how they wrote it, you can have dramatic
difference in speed, several orders of magnitude, on the same machine.

I asked how much speed difference is important to you. No answer. I asked
how important reliability and uptime are. No answer. I asked if you
considered O.S. alternatives. No answer. Same with all the questions I
asked. These and other questions are very relevant, and speed may be the
least of your concerns.

Even if speed is a concern, it likely will make the least difference which
computer runs the application. You could get thousands of times faster speed
by rewriting an inefficient algorithm, at best some fractional improvement by
picking a different CPU.

You tell your I.T. department that initial cost ia only a fraction of Total
Cost of Ownership (TCO), and to start thinking with their big brains.

 

[=?ISO-8859-1?Q?Arne_Vajh=F8j?=]

Sorry for the approximations about the description of the devices, but
i am not the IT.
Today, my application runs well on a SOLARIS 5.9 V240, and the IT
proposed me to either buy a new V240 or to get a linux/opteron.

I need to use the CPU for XML processing, that is the bottleneck of my
application (I have seen the linux solution is 2.8GHz, 2x2 procs...
therefore I though it was really faster).

Before sending my answer, I want to know this group point of view

PROS:
faster => I need CPU to highly process XML
cheaper => IT said it is 2 times cheaper

CONS:
migration

http://www.spec.org/cpu2000/results/cpu2000.html

Sun Fire V240 (1503MHz)
SPECint_rate2000 18.4

You do not identify the Opteron, but if it is a 290, then:

Fujitsu Siemens Computers PRIMERGY BX630, AMD Opteron (TM) 290
SPECfp_rate2000 79.7

Migration will cost nothing because you are using Java.

RHEL is an excellent server OS.

Four times the CPU power for half the price.

I would not be in doubt !

Arne

 

[Guest]

The application uses a lot more than the CPU. The CPU is likely not the
bottleneck at all.

It is very common for XML processing apps to be CPU bound.

XML processing is subject to inefficiency in the software. Depending on
what software you use, who wrote it and how they wrote it, you can have
dramatic difference in speed, several orders of magnitude, on the same
machine.

Very few people write their own XML libraries in Java.

Arne