At Last, Windows DataCentre Edition
Ok,ok, I'm outa here.
Mainframe hardware got a refresh at the high end last year and in the midrange this year, but that is it for this level of big iron from IBM for a while. However there are plenty of enhancements the company will add to its mainframe operating systems to keep its largest customers happy – and therefore happy enough to pay the …
IBM claims they can consolidate 1.500 of the x86 servers on their biggest Mainframe. After some digging, It turned out that all x86 idle and the Mainframe is 100% loaded! And because a z196 Mainframe cpu at 5.26GHz and more than 100MB cpu cache - is slower than a decent x86 server cpu - how can 24 of the z196 cpus consolidate 1.500 much faster x86 cpus? I dont get it. What happens if a single x86 cpu starts to do some heavy work? Then the Mainframe cpu can not keep up.
And now IBM talks about "100.000 virtual machines". Great. Are all those VMs powered off? Are they even installed? Or are the VMs old antique computers, such as C64? Or 486 cpus at 25 MHz?
There is no way a Mainframe can cover up 1.500 x86 servers. Even less 100.000 VMs. If you have 100.000 of the 486 cpus @ 25MHz - then you get quite hefty cpu performance. That no Mainframe can cover.
Pure IBM Marketing. Again.
Again, if you just thought a bit over the IBM claim? Instead of believing the IBM marketing, ask yourself some basic questions. When you did an exam test at the University, have you not been taught to ask yourself if your answer is reasonable? When I hear something, I always critically analyze the statement: can it be true? Is it reasonable? No matter what I hear, I always do that. Too much math studies, maybe.
I immediately asked myself:
Is it really reasonable that a (slow) mainframe and 4 racks can virtualize 100.000 machines?
How many x86 Intel Xeon servers would it take to virtualize 100.000 machines? Let us be generous and assume that one 4U Xeon server can virtualize 100 virtual machines. Then, to virtualize 100.000 machines, you need 1.000 Xeon servers. Do you really think that 1.000 Xeon servers can fit into 4 racks?
Have you never heard about quick and dirty back-of-the-envelope calculations? You never question anything you hear?
"...Pure Kebabbert Marketing. Again. You are no better than IBM...."
Oracle/Sun keeps claiming 14K thread support for Solaris and how it makes it a generation ahead of every other O/S.
The problem is the most threads on any hardware that supports solaris is 512 which is half of what AIX can do. And who is going to buy a 795 with 256 cores and only run one operating systems on it?
People who live in glass houses should not throw rocks, and should read the actual claim from IBM which was about their blade attachment not just the mainframe.
Pure Kebob again
Actually it is 16K threads, not 14. Have you heard about the number 16.384? It should ringa a bell, if you are a technical person.
"...The problem is the most threads on any hardware that supports solaris is 512 which is half of what AIX can do...."
You are not making sense when you are comparing the old Solaris 10 to the newest rewritten AIX. You should instead compare the old previous generation 512 thread Solaris 10 servers, to the corresponding POWER6 servers. The biggest POWER6 server, the P595 had 32 cpus for a total of 64 cores. The POWER6 runs only two threads, because it has two cores.
Thus, compare the previous generation Solaris servers with 512 threads to the previous POWER6 servers with 64 threads. We see that IBM POWER servers have scalability problems. And what is your conclusion? Quite the opposite!
And Solaris 11 (which will be released in 2-3 weeks) had a rewrite, just like AIX, to increase scalability. But when AIX increaed scalability from 64 cores/threads to 1024 threads - at the same time - Solaris increased from 512 threads to 16.384 threads.
So... is Solaris a generation ahead or after? What is your conclusion? What new hot tech has AIX invented lately that everyone wants? No ZFS, no DTrace, no Containers (WPAR), no nothing. Bad scalability too. Where are the single core POWER cpus at 7-8GHz, because "databases runs best on strong cores"? And IBM is now selling a container full of servers, just like the old Sun BlackBox. As HP does. IBM is copying everything that Sun did. And now IBM is lowering their prices - which is a sign of weakness. And Oracle is increasing prices - which is a sign of strength. Not to mention when x86 will be faster and cheaper than POWER cpus, and soon after AIX will be replaced with Linux. Better learn Linux, boys, so you can keep your job when IBM announces end of line of AIX and POWER servers. IBM is a generation behind. Nothing cool comes from IBM, just pure marketing that everyone here believes, except those with a brain and can ask themselves basic questions.
"....People who live in glass houses should not throw rocks..." - I can certainly not agree more.
That is a lot of worda to avoid admitting that Oracle/Sun keeps claiming 16,348 thread support for Solaris yet the most threads on any hardware that supports solaris is 512 which is half of what AIX can do. And, all I heard this month at OOW was how Oracle finally has single thread performance but only on the four socket box.
You bore us with your essays and we are not naive about oracle marketing
Ok, we'll see if Oracle lies about 16.384 thread servers later.
Actually, I have no problem with Oracle marketing. The reason I have no problem, is that Oracle does not lie nor FUD. Sure Oracle can say that you get 1 million USD if you succeed in getting this SQL query to complete in 100 hours, when OracleDB completes it in 1hour - but that is not a lie. It is spectacular marketing, but not lies. It will not happen that someone did this, and Oracle refused to pay out 1 million dollars - that would be a lie.
If Oracle says that they will beat IBM in TPC-C, and if Oracle really beats IBM - then it is no lie.
Can you give an example of Oracle advertising where Oracle lied?
Regarding IBM, there are lot of examples where IBM lies. For instance, do you really believe that 1.000 x86 4U servers fit into 4 racks? Do you really believe "the z196 Mainframe is worlds fastest cpu"? I have lot of examples, you want to see them? IBM lies all the time. Maybe Oracle lies, but not all the time. Can you show me an example of an Oracle lie?
Sure Alisons comparison with regards to threads is fair. He is comparing current shipping products to shipping products. Now you saying that this will be changed in a few months is also fair.
As for the 100.000 virtual servers, I have no doubt that a mainframe could do that, but hey is it relevant?
I guess not just as relevant as having a single (non supercomputer) OS image using 16.384 threads.
And as for Oracle FUD and marketing lies.. do you really want a list ?
"...Sure Alisons comparison with regards to threads is fair. He is comparing current shipping products to shipping products..."
We had this discussion before, and you know that I am of the opinion that you should current generation of tech, to a competitors current generation of tech. For instance, it is not fair to compare a new tech to an old tech, like comparing POWER7 to Niagara T2.
Companies release their products not synced with each other. For instance, Company A may release product X. Some time later, company B may release product Y, where Y replaces five year old product Z. Then you should compare X vs Y. You should not compare Y vs five year old Z.
Thus, Niagara T2 should be compared to the contemporary POWER6. Niagara T3 should be compared to POWER7. Niagara T4 should be compared to the coming POWER7+.
It is not fair to compare scalability of old Solaris 10 to the latest AIX. Instead, compare latest AIX to Solaris 11, which will be released in two weeks.
"...As for the 100.000 virtual servers, I have no doubt that a mainframe could do that, but hey is it relevant?..."
I have serious doubt. As we all know, a Z196 Mainframe cpu, is much slower than a high end x86 cpu. How can 24 Mainframe cpus replace 100.000 much faster x86 cpus? It does not add up, does it?
The only way I can see 24 slow cpus replace 100.000 much faster cpus, is if the faster cpus all idle, or are powered off. But then it is not really valid statement; who is interested in buying a server that can replace 100.000 shut down x86 servers? No one. That would be of no use.
I can startup several Mainframes on my laptop, via the emulator TurboHercules - this is true. But it is a stretch to say that my laptop can replace 10 IBM Mainframes - which is not true. Who is interested in buying a laptop which can replace 10 Mainframes if the Mainframes are not allowed to do any work? That would be a totally useless laptop.
"...And as for Oracle FUD and marketing lies.. do you really want a list ?..."
Not a long list, but a few examples does suffice. For instance, if Oracle promises to beat IBM in TPC-C, and Oracle really beats IBM - it is not a lie nor FUD. It is the truth.
If Oracle promised to beat IBM in TPC-C and did not beat IBM - then it is a lie and FUD. Then it is not a true statement.
FUD according to wikipedia, is if you deliberately say something false, twisting it to make it sound true. In short; spreading false and evil rumours, to discredit a competitor.
"...I guess not just as relevant as having a single (non supercomputer) OS image using 16.384 threads...."
Well we will see later if Oracle really does release a 16K thread server, or not. Yes, it sounds crazy today, but so did a 8-core cpu sound crazy back then. Now everybody has 8 cores.
IBM mainframes are like a container ship, there are not very fast or very flexible they are just able to run a large number of applications, none of them very fast. Mainframes are ok for batch just don't try anything real-time on them.
I've been involved in several functional migrations of systems such as MQ from mainframes because they were an inflexible single point of failure.
"...IBM mainframes are like a container ship, there are not very fast or very flexible they are just able to run a large number of applications, none of them very fast. Mainframes are ok for batch just don't try anything real-time on them..."
At last a person that have used Mainframes and can attest how slow they are. Good riddance.
Biting the hand that feeds IT © 1998–2019