Originally posted by kraftman
View Post
Originally posted by kraftman
View Post
Originally posted by kraftman
View Post
Originally posted by kraftman
View Post
But the SPECjbb2005 benchmark (configured with multiple JVMs) is one of the problems of the "embarrassingly parallel" class. You can scale it by running multiple copies. In the bench, they configured 128 JVMs, exactly the number of blades. The application didnt have to use NUMA link between the blades. The JVMs are working independent from each other, so no communication between the the nodes, too. So the SPECjbb2005 benchmark on an Altix 4700 is not much more than throwing a cluster at a problem it's best at! Hence, this does not prove anything about scalability.
We all agree (including Linux scalability experts), Linux scales well on a cluster. That is where Linux strength is, horizontal scaling. As Linux scalability expert Greenblatt says:
"Greenblatt: Linux has not lagged behind in scalability, [but] some [Unix] vendors do not want the world to think about Linux as scalable. The fact that Google runs 10,000 Intel processors as a single image is a testament to [Linux's] horizontal scaling".
"The true Linux value is horizontal scaling."
To my knowledge, there is no single big machine with 10.000 intel processors on the market. Google must be using clusters.
Originally posted by kraftman
View Post
To me, a paper is good if it uses official standardized benchmarks, in a fair way, where no one complains. The outcome does not decide if a white paper is good or not. If the white paper has a scientific, reproducible approach, then it is good. Just like in science. The method is extremely important, the outcome is not important in science.
But you have proved earlier that the outcome decides if a white paper/benchmark is good, not how good the method is. I think you(?) showed earlier that one guy migrated from 800MHz SPARC to 2.4GHz Intel dual core Linux and therefore you "proved" that Linux is faster. That is just a weird comparison according to me. Not fair. But that doesnt matter to you, as you have "proved" what you want.
I want more benches on the SAME hardware. With many CPUs/cores. But I suspect you dont want such benchmarks, because Linux would loose big time. You prefer unfair benches. That is not fair.
Regarding Solaris scaling to 512 CPUs, I dont know if there exist such a Sun machine on the market, so I can not show you such benches. But if there existed such a machine, Solaris would of course have higher CPU utiilization, again. Why wouldnt Solaris?
You know, just because Linux EXISTS on a big machine, it doesnt mean that Linux scales well. I bet that Linux has a very low utilization on machines with many CPUs, maybe 10% CPU utilization? But I am sure that Solaris utilizes the CPUs far better than Linux on same hardware (machines with many CPUs). As we saw on the SAP benchmark with 48 cores.
Originally posted by kraftman
View Post
Originally posted by kraftman
View Post
Comment