Sun Bashes Linux on (IBM) Mainframes

dagbrown writes: "An article linked from Sun's front page, entitled "Linux on the mainframe: Not a good idea" by Shahin Khan, Sun's chief competitive officer, has the interesting theory that Linux on mainframes makes no sense because, among other things, the VM/Linux combo isn't a very good match. What do the folks on Slashdot think?"

Maybe not,

[Cardhore]

But free software on a mainframe isn't bad. Remeber, we also have such things as "FreeBSD", "OpenBSD"; also "NetBSD." Yes, they're new to me too.

Of Course not!

[Dragonshed]

Why would it be a good idea when you can buy Sun machines with Solaris, and get much more quality computing power for your dollar!

All this is is Sun actively protecting their brand.

my 0.025$

I actually worked on Linux on a Mainframe...

[Dixie_Flatline]

And it was a serious pain in the ass. There were problems with the virtual machine suddenly giving up the ghost from underneath us, and we'd see Samba processes go wild for no reason whatsoever. We had load averages spike well into the hundreds, and it was like we were always scrambling to keep it running, as opposed to setting it up and just having it work. We used to tell the students that the machine had caught on fire and had (literally) fallen over. We were even thinking of doing up artwork.

All those impressive demos where they have 32 hojillion instances of linux running on a mainframe are meaningless. Sure, you can do it, but it doesn't do anything. If you try actually working with the setup, you'll be rebooting your machine 10 times a day, and those mothers take forever to freakin' reboot.

IBM milking mainframe monopoly

[ChrisRijk]

Just a little FYI:

With Amdahl checking out of the mainframe business it seems IBM has decided to raise mainframe prices significantly - it's actually charged more for the same performance in 2001 than in 2000! This is why IBM's mainframe revenues increased by a fair bit between 2000 and 2001 (while it's PC and Unix revenue dropped). Mainframe revenue accounts for about half of IBM's total server revenue...

Benefits of the mainframe

[kick_in_the_eye]

The benefits of the mainframe are plenty. Awsome uptime, its in years, not months. Amazing I/O and storage capabilities. ESCON is unbelievable in the way it works, sharing I/O through timeslicing it. Time slicing CPUs. What you can do on the new Sunfires (Ex800 and E15K) are pretty minimal compared to the slicing and dicing of mainframe. Its more a story of hardware than software. The best solution would be Solaris on the mainframe :-)

Hands-on experience with Linux on a mainframe

[ciurana]

I agree mostly with the article because I recently evaluated an IBM mainframe against an AIX SP2 and a Solaris 4-processor server. Most of the issues in the article, particularly performance, are right on.

The application we were testing was extremely processor and memory intensive. While there was a web component, the biggest problem was moving a large number of bitmaps in one format into the server, convert them from base 64 to a binary representation, rasterize them, and convert them to a "browser friendly" format such as JPEG, GIF, or PNG. We had to complete hundreds (> 200) of these operations per second.

I really wanted to use Linux because most of my staff is familiar with it and our customer felt warm and fuzzy about using IBM equipment. At the end of the day, however, the Linux mainframe only gave us 25% of the minimum speed that we needed for our process to be successful. IBM and a certain German Linux company tweaked everything they could but the performance wasn't there. The AIX vs. Solaris match was more evenly paired. My customer decided on Solaris because they offered a few advantages in Java tools that AIX didn't have. All vendor's boxes had equivalent processor and memory configurations.

I would like to spread the Linux credo as far and wide as possible. What we must understand is that, in order to make Linux a viable option in mission-critical applications (the kind of thing sitting on a mainframe), the performance and "hardening" of something like MVS must be present. Linux just isn't there yet.

Disclaimer: I'm under NDA so that's why some aspects of this posting are a bit vague. Drop me an email if you want more details regarding our experience but our conversation will be "off the record."

Have a nice wknd,

E

Re:Can't blame em...

[sxpert]

ok, I have to answer this bull alltoguether

I'm an IBM tech (not speaking for IBM, of course) and can tell all that the "proprietary Operating system" they are talking about is not really an OS, more like a virtual machine that handles all other oses. This virtual machine is n implementation of the S390 architecture that will handle several "virtual computers" at the same time by the use of multiple "virtual processors", each one used for a "virtual computer", with each having it's own PC, virtual memory handling and all.

Re:Misrepresented article..

[sxpert]

Although z/VM can start and stop Linux images, it cannot dynamically add resources to match demand. As a result, a mainframe would need to size for peak demand just as the Linux farm would; high utilization is a myth.

This is total bullshit, check the IBM RedBooks on z/VM

Re:I disagree

[sbuckhopper]

Say, like an Sun E10000 or Sun E15000 which costs a ton more than that IBM server quoted (for about the same power)...oh yeah, and Sun won't let you buy one unless you order a service contract...meaning, they come with a repair man.

Linux not ready for Mainframes

[rlangis]

Wasn't there a story not too long ago that mentioned how Sun was going to support Linux on lower-end machines, but NOT on the high-end Enterprise [sun.com] systems? (bah, I can't find the link) Anyway, people were saying "Well, Linux isn't ready for Enterprise-type systems yet, so keeping the proprietary *nices on these systems isn't a big deal."

Now, Sun comes right out and says this, and people start complaining? Sure, perhaps Sun is trolling for /. Yeah, right.

You may think I'm biased: I work for Sun, after all. Don't get me wrong - I'd absolutely *love* to take one of the *THIRTY* E10k's I have sitting around me at the moment and install Linux on it. Or, rather - I'd love to TRY. But I don't have any real notion that any version of Linux, AS IT STANDS RIGHT NOW, will work as well as Solaris on that box.

Sure, Solaris isn't very user-friendly. GNU/Solaris (Solaris with GNU Tools) is better, but still not anywhere near what most Linux folks are used to when it comes to command-line fun. However, Solaris is *made* to work with Sun hardware. And it does, very well.

I doubt it highly that someone is going to go buy a US$4M E10k/E15k box and start porting Solaris tools and system utilities *just* so people can run Linux on those systems. Right now, the only reason people have installed Linux at ALL on those systems is for bragging rights.

If you want to outlay the cash and start-a-porting, I applaud you. I really do. But I won't hold my breath.

Re:Makes sense to me

[Steveftoth]

Do you even know what kind of VM IBM is talking about when they talk about their zSeries?
I'll use tech that you are probably familar with. It's exactly like VMWARE for windows/linux. Except that the zSeries OS that runs the virtual machines runs at a much lower level then the VMWARE program and the HW is optimized for the execution of VMs. Unlike the x86 on x86 emulator that VMWARE does, the zSeries boxes run their VM's very fast. All the code that is executed runs native to the processors in the box, you have to recompile everything to run on them.
see this url for more facts on IBM's stuff. I'm not saying it's the answer, but this paper from SUN is FUD.
http://www-1.ibm.com/servers/eserver/zseries /os/li nux/facts.html

Guy's got his facts wrong.

[Blrfl]

NitsujTPU writes:

If the article is correct, then the z800's running zVM emulate Intel x86 architecture in order to run Linux. Heck, even poorly written native compiled code generally has advantages over such a set up.

That's not what IBM is doing at all. The version of Linux that runs on the 390 platform is natively compiled for that system and has been tuned to work with it. This is no different than Linux on any of the dozen or so other platforms it's been designed to run on.

Sun's FUD-slinger got his facts wrong more than half the time in that article, which in most places would get him grade of "F". I've been using and buying Suns for 16 years, and while I really like their products, I hope IBM takes every opportunity to point out what a nitwit he is.

The moral: Just 'cause it says "Linux" on the label doesn't mean it's running on an x86.

Goes to show...

[clump]

Linux isn't designed to run in a virtual machine;

Linux isn't designed at all, which is good. Thats why its so flexible. There was already that debate [slashdot.org] a while back.

Applications that run on Linux for Intel need to be recompiled and recertified for each new platform; thus the application portfolio to run Linux on a mainframe is small(9).

Oh please. Like your going to have an easier time compiling non-Linux software? Still think so given how open and portable most Linux software is? Is mainframe software as portable? Is there lots of free mainframe software to port? Thats almost as irrational as Microsoft's "Linux isn't free" TCO argument. Per that can-of-worms, because both systems have TCOs means NT itself *is* free?

Articles like this are interesting because Sun definitly has a conflict of interest with Linux. They need to appear as if they support it so new blood will buy SPARC hardware with Solaris, but they also don't want people 'liking' Linux over Solaris/SPARC.

Personally, I love Linux on SPARC. I would prefer Sun making Linux more 'Enterprise'-like instead of hawking Solaris as a big-brother. However, I understand that Solaris is a huge investment and one they probably will think is superior for years to come.

For their sake, I hope the Penguins don't squish them. But if they don't look both ways before crossing the street...

z/VM is *not* x86 emulation...

[Noel]

People these days seem to forget about the overhead of interpretters and virtual machines. If the article is correct, then the z800's running zVM emulate Intel x86 architecture in order to run Linux.

Sorry, but z/VM has nothing to do with emulation. z/VM is a low-level system that simply (or not so simply ;-) virtualizes the hardware by providing one or more virtual machines, each of which can run any native OS. As far as the client OS knows, it's running on the bare hardware. The z/VM layer provides the ability to flexibly divide the hardware resources between the VMs, and guarantees that each VM is completely isolated from all other VMs. In the case of Linux/390, the Linux kernel and applications have been compiled to run natively on the S/390 architecture. Check out this Linux for S/390 FAQ [marist.edu] for more info.

Re:Sun is not Linux's friend

[endemic]

Sun sure seemed to be Linux's friend about a week ago while they were flaunting Linux on their new x86 server line.

Check out the Sun Broadens Support for Linux [sun.com] feature.

Re:Sun is not Linux's friend

[arunkv]

From a Linux advocates point of view, there isn't much difference between Sun and Microsoft. Don't be fooled by the saying "My enemy's enemy is my friend", because it doesn't apply here.

If you look at Sun's home page, the article just below the one bashing Linux on the mainframe is:
Feature Story: Sun Broadens Support for Linux [sun.com] Aggressive new program expands the role of Linux on entry- level servers.
Go figure!

Re:It's FUDerific!!

[LoseNotLooseGuy]

Sun is loosing a great deal of business recently due to the introduction of Linux into the server market.

I find that unlikely. I think Sun's stockholders would be very upset if Sun were actually "letting loose or releasing" business. However, it could be argued that Sun is failing to retain business due to Linux deployment. The word you were looking for is losing.

Congratulations! You have been participant #35 in my campaign to rid Slashdot of this error.

Partitioning != VM

[ansible]

System partitioning isn't the same as IBM's VM technology.

With your E15K, you're dividing up processors and memory between various partitions, each running an instance of the OS.

IBM's running multiple OSs as virtual machines on the same system.

With Sun, if partition A is really busy, and partition B is idle, you can't make use of those idle processors unless you re-allocate your partitions.

With IBM, the processes in VM1 can use all the processors of the mainframe, unless VM2 also needs processing time.

With either one, if one OS instance crashes, it shouldn't affect the other instances.

Re:I disagree

[plankers]

Older versions of AIX, namely AIX 4.1 and 4.2, weren't very good, standards-wise. A lot of this has been fixed. IBM's latest version of AIX, AIX 5L 5.1, includes a lot of stuff they got through Project Monterey. Project Monterey was the combined effort between SCO, IBM, and Intel. When they joined, IBM had access to the SysV source for the first time, and started making their OSes more SysV-compliant. This was coupled with a general rewriting of all the device drivers in the OS, so that the drivers were more robust and handled errors better. If you watch carefully, you can see the effect these efforts have had on AIX, even in AIX 4.3.3. AIX 4.3.3 has had some interesting, destabilizing bugs in the last year. They were patched rapidly (patched, but IBM Support always had a good workaround if you couldn't just reject the faulty software package and put the machine back the way it was -- AIX has excellent package management!). Looking at the bugs, though, they were very low-level bugs that seemed to get exposed by development work throughout the OS. Given the sorts of things that are changing in AIX, it is actually a good sign that deep-seated problems are boiling to the surface, because that means that the developers are really doing their job.

The other thing that seems to have improved AIX 4.3.3 a lot is the benchmarking. Dueling with Sun isn't a bad thing all the time. AIX 4.3.3, with the latest patches applied, has an excellent, capable, and very tunable TCP stack. It has also had a number of features backported to it from AIX 5.1. AIX 5L 5.1 has a lot more of the cool TCP features in it, and is even with Solaris 8 in those regards.

In the past, compiling has seriously sucked under AIX. However, the IBM VisualAge C++ compilers for AIX are cool. They will compile just about anything, and if you cannot compile it then the programmer should probably go back and adhere to the various C/C++ standards. Generally I have the same trouble compiling certain software under either Solaris or AIX. Sometimes I even break down and install the GNU compiler collection...

Right now, IBM has two machines that support virtual machines, or in mainframe/IBM terminology, logical partitions (LPARS). LPARs are supported in hardware on the pSeries 690 running AIX, and on IBM zSeries running z/OS or z/VM. LPARs are hardware-based, and if a processor, memory chip, or other system component dies it is able to prolong an outage by deallocating it. Currently AIX is not able to dynamically add resources to an LPAR without rebooting the LPAR, but that is coming. Otherwise, LPARs are totally independent instances of the OS. And the reboot to add resources takes an LPAR about thirty seconds (yes, 30). The LPARs do share a few things, like the system clock, and a management workstation for console access, but they are isolated in hardware.

On the pSeries 690, the boundaries of the LPARs are 1 processor/1 GB memory. So you have to allocate whole processors and whole gigabytes of RAM to an LPAR. While this sounds like it's not such a great idea (and it isn't the best), it's okay. Any more granularity can be handled by the built-in AIX Workload Manager, which is able to manage OS-level resources like memory, processor time, disk I/Os, etc. in percentages based on users, groups, and process names. That isn't something Sun has, and often you don't need partitions on a machine, but instead just need to keep two pieces of software from fighting for resources (or need to cap a group of users to low memory or low CPU usage). Workload Manager is very handy in that regard. The IBM mainframes running z/OS or MVS can do timeslicing, where an LPAR can have 10% of all of the CPU time on the box, or 5% of RAM, etc. You can also create situations where you overcommit the resources on the machine but define priorities, to guarantee levels of service. So maybe you have three LPARs, and one can have up to 80% of the CPU if it's busy, and the other two can have up to 30% of the CPU, but LPAR 1 has priority (so it gets its 80% anytime it needs it). This is where the pSeries 690 is going, it just isn't there yet (IBM took all the guys that made MVS capable of this stuff and pointed them at AIX).

I really know nothing about Sun's Dynamic System Domains. I do know it is similar in certain ways to the concept of LPARs, but isn't as flexible as the MVS/mainframe LPAR scheme. The OS instances are still isolated from each other in hardware. The two are probably nearly even when it comes down to it.

If you consider the hardware these things run on you will see that a pSeries 690 with 32 CPUs equals a Sun Starfire with 72 CPUs. What IBM is doing in this regard is cool -- they are actively attempting to put fewer components into a machine. More components == more likelihood of failure, and therefore you need more redundant components. More components == more heat, which leads to more failures. It also means more electricity for more components. You can also cluster the big machines using a derivative of the IBM SP2 technology (you can also cluster smaller ones, like pSeries 660s, which are a hell of a deal, price/performance-wise). So while Sun's machines can go to 72 CPUs, all you get are extra components and extra heat (and extra service calls)

IBM has been working with Red Hat for a while now. In fact, when Linux was ported to the S/390, there were two ports. I forget who the individual was that did one of the ports, but IBM had a team that did the other. There is an S/390 distribution of Red Hat Linux -- check any mirror site that is worthy (many don't carry it). It generally lags behind the i386 distribution, but only by a few weeks. Considering the market, hey, that's not bad at all. And IBM is doing a lot of work on Linux itself, from porting JFS to it, to adding a lot of what makes AIX a very scalable, very stable, very reliable OS. Check the IBM AlphaWorks site for examples. I am not sure what Sun is doing in that regards, or what Solaris can offer Linux as far as technology. That isn't an insult, it is just my unfamiliarity with Solaris.

And finally, every machine is stable & reliable with a good system administration team that is knowledgable. I have been doing AIX admin work for about two years now, and I love it (having come from a Linux admin background). AIX has always had excellent filesystem support (true logical volume management built right in with real journalling filesystems, while you have to buy it from Veritas for Suns), and that, for me, makes it very easy to work with. And if you work with an OS for a while you get to know how it does things. I cannot speak for the admins that deal with Suns. I don't work with Suns, but half of the machines where I work are Sun machines (the other half is AIX, with two mainframes, and a handful of Linux boxes). And those Sun sysadmins can make their machines as stable and reliable as my AIX boxes. So it just depends on what you're trying to do, what you already have, and often, how much it'll cost you, because more often than not the machines can do the exact same thing, except one will be thousands cheaper.