How it's possible you don't know Linux scales to 32 cores while you show paper where it scaled to 48cores and I showed you it scaled linearly up to 64 cores? Idiot? :>In one word: Ridiculous. It doesnt matter if this discussion is eight years old, it only shows the Linux SCALING EXPERT's attitude. They claim Linux scales to 10.000 computers, and at the same time they say Linux scales to 16 cores - which is excellent scaling!!! Of course Linux scales better today. It may even scale to... 32 cores? I dont know. When Linux scaling experts today say that Linux scales excellent, they probably mean it scales to 32 cores today. But the step from 16 cores to 4096 cpus, is... quite a big step and not very likely? IBM highend Unix AIX does not scale well even today, after decades. It takes decades to scale well. And the developers need to have access to big machines. Linux developers mostly use 1-2 cpus with quad cores. I really do think Linux are fastest on such desktops, faster than Solaris. But they typically dont have access to large machines, and can not tune Linux to such machines.
What's wrong here? Like stated hardware can limit operating system to achieve more utilization.You know what kind of servers are expensive? The servers with high RAS are expensive. You pay for RAS. Not performance. Performance is not expensive. IBM Mainframes are dog slow cpu wise, but have very good RAS. That is the reason IBM Mainframes are shitload expensive. Wrong again.
And, Solaris machine used more RAM, different hardware and database. Where did you get those revelations about RAM sticks from? It doesn't change the thing if SAP benchmark requires 256GB RAM, but it can have influence on scalability.And, Linux machine used faster RAM sticks and faster CPU. The reason Linux machine did not use 256GB RAM, but instead went for the lower 128GB RAM, is because if Linux went for 256GB, then Linux must use much slower RAM sticks. I have written this many times. But you just ignore it. SAP benchmark does not require 256GB RAM, because then Linux would go for 256GB RAM as well.
Will that mean scalability isn't affected by amount of RAM? Sounds ridiculous.If you wish, I can mail SAP technical support and ask how many GB RAM is required to complete SAP benchmark. Maybe it is 8GB RAM? Or even 12GB? Or maybe it is better that you do it. Because when I show you the answer from SAP, you will just say "lies, it is FUD, it is Sun propaganda".
I think you're trying to undermine SAS benchmark which shows Linux reached 95% of CPU utilization, scaled wonderfully and it didn't even reach higher utilization, because of hardware limitations. I wonder if any OS could handle this workload and I wonder Solaris could reach 95% of CPU utilization on a 64 core machine. :>The SAS benchmark simulates many users that log on and do some simple tasks. This work load is easy to parallellize. Just like SETI client. And the work load has very low CPU utilization. This is a workload which is not taxing nor demanding. Any OS could handle this work load. For instance, let 5 independent users be assigned to one core. When there are 10 users, assing two cores. etc. Does this work load sound challenging to you? Does it sound like hard to parallellize? Do you think SETI is hard to parallellize?
I'm failing to see how this 64 core machine isn't a single system-image big iron, where _one_ kernel runs on a single machine. I also don't understand why did you paste this. How is it related? Pasting more and more text won't help you in any way.We had this discussion before and llama wrote about scalability and work loads that are hard to scale (SAS benchmark is not hard to scale). You was in this thread, but still you ignore it.
"We're not talking about clusters. We're talking about single system-image big iron, where _one_ kernel runs on a single machine with > 16 CPUs in a cache-coherent shared-memory system. The most cost-effective machines for cluster-building, in CPU power per dollar, are dual-socket quad core Intel Core2-based machines. i.e. 8 cores per node. That's great if you have a workload that has some coarse-grained parallelism, or is embarrassingly parallel, e.g. processing 100 separate data sets with single-thread processes that don't depend on each other. That's not so great if you have a lot of processes that need fine-grained access to the same shared resource. The canonical example here is a database server handling a database with a significant amount of write accesses. Otherwise you could just replicate it to a big cluster and spread the read load around. Locking for write access in a big cluster, even with low latency interconnects like infiniband, is still _way_ higher overhead than you'd get in e.g. a 4 or 8 socket quad-core machine. Even NUMA big iron is better suited for this than a cluster.
It's funny you're jumping at clusters right now.CLUSTERS DON'T COUNT AS BIG IRON. They're just a pile of normal machines. They do have their uses, though."
You showed some crap, unrelated papers and you spread the FUD. It doesn't matter if you showed some crap and unrelated papers. It's such simple.I showed my post with all research papers several times to you. And each time you denied they exists, that I FUD and lie. And you ask me to post research papers, otherwise I FUD and lie. And I post them, and you still say I FUD and lie:
But I said the paper was there all the time.
In short I have proved you FUD and lie about me. The paper and the rest of the research papers are there. You just deny them. Obviously, you are trolling, Kraftman. Just as usual. You also confessed you FUD, Kraftman, in a post. Here I show you lie and FUD about me, and about Solaris. Earlier you confess you FUD. this is fun. How much longer will you claim that it is kebabbert that FUDs? I have showed links and papers and research papers, and links to Linux kernel developers. There is no lies there, nor FUD. You are quite stubborn. Dont understand when you have lost a battle. Sure, continue to say that Solaris scales to 64 cpus, but that is not true. Continue to say that I did not post research papers, that is also not true. etc.
I explained this, but you're ignoring this as a troll like usual. How many times I have to repeat what I said to such troll accept this? The whole point is you're basing on two different SAP benchmarks and you want me to believe Solaris scales better. This is damn stupid.Of course Solaris does not scale. You have explained to us that 128 cpu Solaris machines such as this:
does not exist. That we lie and FUD. And also, the Solaris machine with 144 cpus also doesnt exist. That is also FUD and lies. We are just a bunch of FUDers and liars everyone of us.
I think you should say that Solaris scales to 16 cpus, that is a much lower than 64 cpus. Why do you say Solaris scale to 64 cpus? If you are going to FUD about Solaris, at least FUD properly. Say Solaris scales to 16 cpus instead of 64 cpus.