Dienstag, 13. Dezember 2005

Sun's Niagara and Apache?

Geschrieben von florian um 16:15
Probably I am be the blogger posting the most articles about Sun's systems with the new T1 (aka Niagara) processor without having tried such a system myself, so much that some people call me "Niagara-Blogger".
So, here's another one about the lack of benchmarks with the popular Apache web server. (and in English, too! oh, dear...)

Some people even speculate the good benchmarks Sun published were only possible with the exotic software stack used, including a "special Niagara-optimized Java that nobody outside of Sun gets to use".
So it would be really nice to see some benchmarks with software people can easily download and try to reproduce the results. And since Sun markets the systems as perfect for web serving tasks, why not with the number one web server Apache?

If found only one mention of such a test on the web: Paul Venezia wrote a review of the T2000 in InfoWorld: claiming 2200 pages/sec for the T2000 compared to 2500 pages/sec for an unnamed 4-cpu dual-core Opteron system (8k pages with 10 images on Apache 2.0.54). So that's not bad at all, as I estimate you'd get a price/performance ratio for the T2000 of 75% compared to the Opteron system. However, this benchmark poses some questions: Venezia also states a 2-cpu Xeon system reaches a surprising 2000 pages/sec (with IIS), making you wonder what excactly is measured here. Again, I sorely miss detailed explainations on the benchmark setup.

On the other hand, I hear from a person whose company got to test a T2000 that they just got Apache web serving performance comparable to an 8-cpu E4500, which I find pretty disappointing, since the E4500 is old and the T1 cores run at three times the clock speed.
I don't have any information about the test setup, though (except that it was on some Apache).

So again, there's still no good information out there on how the new T2000 und T1000 will perform with Apache or even a LAMP setup.

On a related note: Sun's MrBenchmark announces a series of benchmark postings comparing quite a few systems, including T2000, and the first benchmark is a FPU-heavy microbenchmark that I assume to be not favorable at all for the T1 processor... this is going to be interesting.
Kommentare (4) | Trackback (1)
Tags für diesen Artikel: , , ,
Artikel mit ähnlichen Themen:
Niagara server for Bloggers?
Sun T1: Finally, a web benchmark, but still no details
Hardware zu verschenken Teil 1
Niagara II
Server-Bescherung
Suns Nikolausgeschenk: UltraSPARC T1
Tip fürs Hardware-RAID auf SUNs V20z
"Sun has an x64 server"
DSL-Modem "heilfönen"
Tags für diesen Artikel: , , ,
Artikel mit ähnlichen Themen:
Niagara server for Bloggers?
Sun T1: Finally, a web benchmark, but still no details
Hardware zu verschenken Teil 1
Niagara II
Server-Bescherung

Trackbacks
Trackback für spezifische URI dieses Eintrags

Sun T1: Finally, a web benchmarks
After I already ranted about missing web application benchmarks results for the new Sun Niagara-based systems apart from SPECweb, finally some results are coming in: Robert Milkowski has some results for a PHP application (with Apache, I assume). Very imp
Weblog: flawed concepts
Aufgenommen: Feb 21, 23:59

Kommentare
Ansicht der Kommentare: (Linear | Verschachtelt)

Florian,

I have also not tried out the T2000 etc... But I was thinking it would be cool to see if it worked out for somebody who would need heavy Floating point muscle.

Thanks for your comments at MrBenchmark's page
#1 Amit Kulkarni (Homepage) am 15.12.2005 14:20
The system is not designed for floating-point intensive tasks. I think the age of general-purpose systems came to an end. It gets more and more difficult to design a system that brings performace advantages by a significant amount on all task. So we will see more and more specialized systems.
This system is a specialized system for a certain task: Highly parallel tasks with a low amount of floating point: To say it short: the internet, transaction processing, databases to some degree. It´s important to know for which tasks you can use this system, and MrBenchmarks page is very valueable to tell the do´s and dont´s for this processor.

@Floran: I´m absolutly sure that the T1 will performe extremly well with the apache ...
#2 c0t0d0s0.org (Homepage) am 21.12.2005 21:46
I forgot something .... there is one exception to the rule of "no fp-intensive tasks". Operations that are important for SSL are accelerated in the CPU. So you get immense SSL-performance without an ssl-accelerator ....
#3 c0t0d0s0.org (Homepage) am 22.12.2005 07:43
@c0t0d0s0:
Yeah, judging from other benchmarks I too expect the T1 perform well on apache. But the numbers I got from a person that actually had one for benchmarks were disappointing to say the least. Now, I don't know how they did the benchmarks and they might have done something wrong in the benchmark setup.
That's why I want to see hard numbers on a benchmark setup everybody can reproduce.

To top it off, I just hear from the same source that the bundled accelerated OpenSSL lib has severe interoperability problems with other SSL implementations.

Yeas, this is all rumour and speculation, but the point of my blog entry was that I really miss some reproducible benchmark results to have some facts here. The published SPECweb results don't really help here.
#4 Florian Laws (Homepage) am 22.12.2005 11:22

Die Kommentarfunktion wurde vom Besitzer dieses Blogs in diesem Eintrag deaktiviert.
 
Powered by Serendipity | Template by Perun