IBM has been hoping for years that Linux would drive new workloads on Power-based systems where OS/400-i platforms are the back-end systems – just as Linux-based partitions have, to a certain extent, been the salvation of the System z mainframe. It is hard to say for sure how much traction Linux has with OS/400 and i shops – or …
@IBM: You Can Scrap Power Right Now
...because x86 will eat the lunch. Even large, wealthy users in the financial industry don't buy Power any more and go the x86 route. Google never bought anything else, but runs the biggest server system of all server systems.
If one x86 box goes down, modern software simply marks this box as broken and processes the remaining workload on one of the 132 other cheap x86 boxes. A week later someone simply replaces the x86 mainboard and the full capacity of 133 servers is again online. All it requires is for software engineers to expect their software to run in huge clusters as opposed to single, inflexible machines.
ssh, rsync and a software engineer is all you need to quickly distribute a server system over hundreds of boxes and make it running. No more expensive IBM hardware required at all.
That is a rather one-sided view
You are basically stating that there is no workload but the cloud load and making a secondary implication that it is number of machines which matter, not task switching within a machine.
I beg to differ.
First of all not all tasks can be parallelised. There are still tasks which need good single thread performance and reliability to complete the task in reasonable amount of time. While x86 has definitely grown up from its humble beginnings there is still enough work for big iron even today.
Second, it is true that we live in a world where a lot of IT admins have grown so incompetent that they cannot be bothered to know particular package dependencies and find it easier to create a VM per app. To add insult to injury they proudly demonstrate the particular memory and resource wastage indoctrination like VMWare, Xen, whatever on the first line of their CV. However even in such a world there are many cases when a machine is shared by multiple apps. In fact, a lot of "apps" internally are multi-process and multi-threaded so your task-switching performance matters quite a lot for your overall system performance.
x86 task switching performance compared to PPC (or even Sparc) task switching performance sucks bricks sidewise through a thin straw. I have a 1.4GHz early Mac PPC G4 which shows lower task switching latency than most Core2 CPUs (at least the ones with a FSB). I would not even try to compare x86 to new PPC hardware with this regard.
So if your 132 machines are doing something that is highly event driven you may find that x86 is actually providing considerably worse value for the money than you think.
I agree with what Anton said, but there's more.
When you buy your 133 Dells/HPs/whatevers, you bought basically commodity parts. So when a mobo goes out in one of those 133 Dells, you call Dell, and they then send out whatever flunky repair guy they have a local service contract with. Not so with IBM. When IBM sends out a guy to work on an AS/400, that dude knows his stuff inside and out. He's not some tech school washout with bad hygiene. And if he hits a problem he doesn't have the answer for, he can call IBM and talk to higher level engineers until the machine is fixed. The Dell guy, meanwhile, is trying to understand the broken English of some screen-reader in India. And you know what? You've still got 132 other computers with potentially crap mobos in them, so you and the Dell flunky may become drinking buddies. Assuming you aren't fired first for wasting such a huge chunk of company money buying a room full of crap servers.
IBM takes their AS/400 customers VERY seriously. Or at least they did when I worked briefly as an AS/400 programmer back in the late '90's.
As an ex-IBMer
>IBM takes their AS/400 customers VERY seriously. Or at least they did when I worked briefly as an AS/400 programmer back in the late '90's.
Not any more! The first step is a phone call to a help line in india, next you may get a visit from a outsourced contractor that may be able to fix your box. And don't talk to me about crap mobos. Odds are the ibm contractor will turn up with a replacement part that is not correct for your hardware. Even when you quote the full model number and part number the stocked items will share the same model and part number but be physically different.
I've had an experience where we had a faulty mobo and requested a replacement model number part number the works! Replacement arrived and it match the failed unit numbers but was physically different. It wasn't able to take the CPUs from the failed unit and a couple of the plugs were different.
All the dudes from ibm who knew their stuff were fired years ago and cheap washouts employed. IBM is driven by managers who only look at costs and not service. Look at what they did to Air New Zealand about a year ago. They had the whole booking system offline for a couple of days at least.
"IBM takes their AS/400 customers VERY seriously. Or at least they did when I worked briefly as an AS/400 programmer back in the late '90's."
And they still do :) We got a brand spanking new Power 7 in our rack and have been having some minor teething problems (got hit by the bug that causes the server to flag an error if the control panel hasn't been used for 25 days, and one of the fans we're having an intermittent error).
IBM sent us a new fan by over night delivery which solved the fan problem, and a rep from IBM has been calling us several times a week just to keep us updated on the control panel bug and to see if it is anything else he can do to help.
I don't think Dell would do that.
In it for the money
Of course the reason IBM want you running Linux is that it hurts their bottom line to develop their plethora of legacy OS baggage. You can hear the shareholders getting giddy already at the prospect of IBM dropping that most unloved AIX.
That said, why would you even consider running Linux on Power? x86/x64 is the only first class citizen in Linux, and only ARM looks to be joining it.
Can anyone think of any commercially available software for Linux on Power? PAVE x86 emulation doesn't count.
I told you so
I said it earlier, and will say it again. Or, actually, I am not telling you anything, but IBM senior executives are telling you:
"Steve Mills, senior vice president of IBM's Software Group, told CNET News.com at last week's LinuxWorld trade show.
Asked whether IBM's eventual goal is to replace AIX with Linux, Mills responded, "It's fairly obvious we're fine with that idea...It's the logical successor."
A replacement "won't happen overnight," Mills said, but years of experience designing operating systems at IBM and other companies means developers know just where Linux needs to go. "The road map is clear. It's an eight-lane highway."
No one believes replacing AIX with Linux could happen quickly, or that IBM will leave its AIX customers in a lurch. But the degree of Mills' Linux support surprised some. "
I told you so. I mean, IBM told you so. This is the first step in killing off AIX, just as IBM publicly said. When POWER8 releases, I expect x86 to be much faster. Then there are not strong reasons to use shitload expensive POWER, when you can use x86 instead.
If IBM really saw a future in AIX, IBM would have ported AIX to x86, because x86 will be faster and much much cheaper. But no, AIX is tied to POWER. And no one can compete with x86 in the long run. POWER will die. When that happens, IBM will kill AIX. Just as IBM said publicly. This is nothing I make up. Yes, IBM said that. It's true. AIX will die.
IBM has said nothing about killing off POWER, and I dont know if IBM is going to kill POWER. That is not sure. But, I find it strange to sell shitload expensive POWER gear, when x86 is much cheaper and faster? So, my guess would be that IBM will kill off POWER when x86 is too far ahead. No customer will be interested then. Or, can you see a reason why POWER would not be killed, whereas AIX is killed off? Any thoughts on POWER?
Not quite right.
There is no x86 that matches the performance of an IBM power6 or power7 chip. Now x86 is cheaper, but it is by no means faster.
We have a power6 box at work to do compiling on. It's one of the fastest powerpc boxes you can get for compiling on, and compiling code on Linux natively is the only sane thing to do. Cross compiling just doesn't work in general. So if you are making embeded powerpc boxes running Linux and you need a build machine, an IBM powerpc box is your only sane choice. They probably make nice database servers too I imagine.
And whoever thought powerpc isn't a first class citizen in the Linux world is seriously mistaken. It is at least as well supported as ARM is.
"....There is no x86 that matches the performance of an IBM power6 or power7 chip. Now x86 is cheaper, but it is by no means faster...."
I suggest you stop read the IBM marketing material and instead rely on real benchmarks.
For instance, IBM claimed that POWER6 had ~220GB/sec bandwidth. If you investigated this further, It turned out that IBM added all bandwidth in the CPU: L1 bandwidth + L2 bandwidth + etc etc. You know, if there is a bottleneck in the cpu at 10GB/sec, then there is no way the cpu will achieve more than 10GB/sec. You can not add bandwidth!
IBM also claimed they can virtualize 1500 x86 servers on a big mainframe. It turned out that all x86 servers idle at 2-3% and the Mainframe is fully loaded at 100%! Would you agree if I claim: "my laptop can virtualize 10 IBM Mainframes with the software emulator TurboHercules (provided they are all idling and just booted up)"? No, you would not agree with this claim. This would be a clear lie.
Regarding your claims that POWER6 is much faster than any x86, look at real benchmarks instead:
"POWER6 is not very impressive despite its crazy specifications. For example, a 4 CPU POWER6 4.2GHz machine ... performs similar to a 2 CPU Nehalem 2.93GHz ... machine in TPC-C. Although POWER6 is older but Intel is closing very fast and that really worries IBM."
"Yes, POWER6 was inefficient in Linpack and similar benchmarks for HPC - roughly 60 per cent achievable out of theoretical peak is really bad for a modern CPU"
Also, on SAP we see that AMD Opterons is faster than 5GHz POWER6
Eight AMD opterons is 2.7x faster than four POWER6. Eight AMD opterons should be twice as fast as four POWER6, not three times faster.
In short, there are numerous links showing that x86 is faster than POWER6.
Regarding POWER7, yes it is a fast and good cpu. But I expect x86 to surpass it soon. AMD is ramping up with 20 cores, Intel is realesing 22nm 10 core cpus, etc etc. Niagara T3 is already faster than POWER7 in some benchmarks (as T3 holds several world records today)
To summarize: Lennart, you are wrong.
#There is no x86 that matches the performance of an IBM power6 or power7 chip. Now x86 is cheaper, but it is by no means faster.#
That is to put a quarter on the lunacy jukebox.
You must be new here.
A Reputable x86 Source
..is definitely NOT Dell, but rather the Local PC Shop You Trust. They know all the components because they repair them daily.
Dell & Co are shipping you the cheapest crap they got in the largest possible volume. Their support strategy is driven by the Middle Finger.
Not so the local guy who knows his trade, products sold,products repaired and customers. But you must indeed be more than a PHB to identify this local guy. How much easier to just "type www.dell.com/shop".
Linux on Power, Mainframes, Itanium and x86 are not the same!!
I am not going to talk about HW RAS or the quality of the support from IBM or any other HW vendor... Think first about the issues of running Linux Red Hat on systems different from x86_64!!
IBM will tell you that Red Hat 6 running on Power and/or Mainframe is the way to open source and everything will be great, but what about the software that runs above the O.S.?? Does IBM mention on any time the term "binary compatibility"?? Go ask your software provider if they will support their application running at Linux RH6 on all these platforms... In order to do such thing they need to recompile, test and maintain the application on a bunch of different platform, which obviously costs a lot more! Just as an example, check when Oracle DB newest version was released for each platform. The last time I checked the latest version for Red Hat running on Mainframe was 10.2, and the versions 11.0, 11.1 and 11.2 were already GA, and guess for which platform these 3 versions were launched first, Red Hat Linux on x86_64!!
We just saw this movie a couple of months ago with Itanium... every software vendor were dropping support for Linux or Windows on Itanium, and then Microsoft and Red Hat announced they would drop the support for the OS as well. Who can guarantee the same thing won't happen for Linux on Power or Mainframe? Sure IBM will support the client, as long it runs Websphere, DB2, Tivoli, etc... but it can't force all the SW vendor to embrace its platforms.
IBM will says anything to get that sale
I buy a lot of Power kit and run AIX on it, linux does not get a look in, AIX and Power go together like SPARC/Solaris, it will always be more stable running the OS it was designed for rather than Linux. Most commentards here have no experience of running Red Hat on P-Series, I don't either, but we do run some Red Hat ("Z-Linux") on the Z-Series mainframes, and it is the most godawful, bug ridden unstable platform in our estate, makes windoze look stable! The constraints with using it on the mainframe are ridiculous, and the resource management/virtualization is terrible, one image can easily go and gobble up all cpu time and starve all other images, we see it ALL the time. IBM also had the nerve to say "Well linux on Z-Series is not very mature, so there will be issues", so why bloody sell it to us then? It's been available for 10 years, is that not mature enough?
Bottom line - IBM will pitch all kinds of bollocks to get a sale, Linux on Z-Series is unfit for purpose, I would bet it won't be any different on P-Series either. If anyone has any expereience of Red Hat on P-Series let me know....I don't think there is many of you out there and for good reason....
And with the nehalem bad boys from intel, why would you want to run linux on Power?? The extra oomph you get from Power is minimal now that x86 has come of age in the high end. The 20-30% increase in processing power you get from POWER does not make up for the 400% extra cost anyway, you might as well just throw hundreds of thousands down the drain...
Perfectly stable on p-series.
We run Debian, and so far other than it being powerpc rather than x86, it is exactly the same. It works, it's stable, it does what it is supposed to do. Of course we are running on bare hardware. I can't give any opinion on IBM's hypervisor stuff on either the p-series on the z-series. But Linux on powerpc is as stable and mature as it is on x86. It just works. Sure a p520 might cost 50 to 100% more than an x3650 with similar performance, but well x86 commodities do help the price of an x86 machine. Can't beat the volumes.
"...We run Debian, and so far other than it being powerpc rather than x86, it is exactly the same. It works, it's stable, ..."
I dont know many Unix sysadmins that would agree that Linux is "stable". Everybody thinks that Linux is slightly more stable than Windows.
Sure Linux is stable. Specially running in a virtualized environment where the device driver part is pretty constant.
I'll call it stable and I've got 15 years of UNIX sysadmin/technical expert under my belt.
- Geek's Guide to Britain Kingston's aviation empire: From industry firsts to Airfix heroes
- Analysis Happy 2nd birthday, Windows 8 and Surface: Anatomy of a disaster
- Review Vulture trails claw across Lenovo's touchy N20p Chromebook
- Adobe spies on readers: EVERY DRM page turn leaked to base over SSL
- Analysis The future health of the internet comes down to ONE simple question…