Sun Bashes Linux on (IBM) Mainframes 519
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?"
I disagree (Score:4, Interesting)
Running Linux on an IBM mainframe doesn't defeat the entire purpose of using open standards like Linux. You still get the man years of free testing, free software, interoperability, and speed. Or rather, IBM gets them. And by tying software you can't charge for to hardware you can, IBM will have come up with a business model for selling Linux systems for incredible sums of money. Quite an ingenious plan - selling Free Software.
Sun's just pissed they didn't think of it first.
Activity and mud throwing may be a good thing. (Score:3, Interesting)
New title? (Score:2, Interesting)
Can't blame em... (Score:5, Interesting)
Sun: A solution looking for a problem
The FUD heard round the world... (Score:4, Interesting)
Now, Sun offers up the ultimate proof: Linux is just fine as long as it impacts the x86 world - but don't dare put it on a platform that affects us.
To be fair, IBM's offering is not perfect - yet. What Sun is preparing for is a future Linux and Big Iron combo that will be. They are afraid, and this FUD is the proof.
Re:I disagree (Score:3, Interesting)
I am pretty sure, whatever kernel IBM chooses to use on their linux mainframes will be THOROUGHLY tested and rigerously tortured and beaten to death until they know exactly how it's going to act. There will probably be a guideline to read for do-it-yourself people on how to make the most stable kernel for the mainframe you have.
Running linux on the mainframe has TONS of advantages, and they will become more pronounced as it gets more popular and more used.
Sun, give me a break, we know you have a big mouth but scaring everyone doesn't work all the time.
Re:Of Course not! (Score:3, Interesting)
Actually IBM's regular mainframes can. When you buy one of the higher end zSeries servers you get a box fully populated with ram and cpu's. If your liscense is for something less than the max # of cpu's and you later need to add capacity all you do is call IBM and they happily take your money and dial into your mainframe, they set a couple registers in the controller board and viola near instant hardware upgrade.
Re:I disagree (Score:4, Interesting)
As opposed to running on a Sun system, which lets you run on a ... oh wait a minute. Those alternatives from Sun are mostly expensive proprietary systems, too, aren't they?
You also get a system that lets you migrate from your existing Linux systems to the IBM system without having to learn the quirks of a different Unix variant. If you use a supported distribution, you don't even have to learn a new distro. And if/when you decide that IBM isn't where you want to be, you can switch to many other hardware platforms with nothing more than a recompile. Sounds like you still get the key freedoms that Free Software is supposed to provide.
Re:I disagree (Score:4, Interesting)
Sun's pissed that they can't run multiple instances of an OS on their E15K systems. You might get Linux running on it, and Solaris is the champ OS on the big Sun machines ... but they are not virtual machine systems. IBM's hardware design lets them run multiple operating systems in parallel on one machine, and even dynamically share processor between them. And then with the VM operating system loaded, you can create multiple virtual machines and run Linux in each one. And VM is very efficient at that. I once ran 6 instances of VM inside itself, nested all the way down. Surely you've heard of the case where IBM tested running 41,000 instances of Linux under VM. It can do that, though that many seems rather pointless. It does let you partition off the resources so you can give each service function you need to run with its own virtual machine, and thus it's own Linux. And on the larger S/390 and zSeries machines, you can even run OS/390 (of MVS legacy) in those virtual machines, and mix/migrate between them all one a single mainframe piece of hardware.
Now personally, I wouldn't do exactly that. That's an awfully big basket with an awful lot of eggs in it. IBM hardware is quite reliable, but not so reliable that you can depend on getting "five nines" on one single machine. Sometimes there are reasons to take the whole machine down. IBM comes from legacy enterprise worksystems and usually early morning Sunday can be scheduled for maintenance purposes. In these days of e-commerce, you don't have such luxury. If you want to be up all the time, you need redundancy, and that machine way back in the corner of the room where "all the servers are gone" isn't redundancy (virtual redundancy, maybe, but you need real redundancy). You need several machines.
That said, there are pluses to IBM's approach as well. If you need to add another class of service, or partition users apart from each other because one needs to do stuff that needs root access? Give them their own Linux virtual machine.
OTOH, well managed, rows and rows of racks with 40 1U servers in each one, running Linux or BSD or NT or W2K or whatever, can be just as effective, if not more so. You can put dual 1.X GHz CPUs in those 40 machines in one rack ... that's 80 CPUs. That's quite a lot. The IBM zSeries can certainly compete, as can the Sun E15K. But those are going to be physically big, and power hungry, machines, too. Take your pick. There's no simple best answer; certainly not for everyone. All this is about is marketing, anyway.
The struggle for hearts and minds (Score:2, Interesting)
There was a stage during the '80s when I was working more as an industry analyst than as a developer when Sun and IBM between them had become two of the then only four serious pillars on which the future of computing rested at a conceptual level.
At that level, Sun was the embodiment of Unix, taking evnagelical responsibility for the cause, and it is reasonable to assume that within their own envorins they genuinely see themselves in that position still.
From my own biased perspective, I felt they abdicated that authority when they allowed their elegant Network-extensible Window System (NeWS) to be rolled by a tide of industry resistance that mobilised against the upstart Sun and behind the then clearly inferior X.
But I'm sure in Sun's hearts they still believe they are the ultimate repository of deep understanding on all things Unix and are being genuine and honest in the technical basis for this critique.
The real problem is thay they can't see beyond their own world view. They do not have places in their heart for deep understanding of either the VM nor the Linux view or the world, let alone the two in combination.
Still Sun struggles to find its own identity and focus, to say nothing of a sustainable business model for the future.
From NFS to RISC, to industrial strength Web servers and on to Java, Sun has been a major contributor to the direction of mainstream computing, but now seems to be edging closer to following the fall to oblivion of that other former pillar of hearts and minds, Digital.
It will be a worse than sad day when we finally have to convey Sun to history, especially if that comes before Java gets to really stand on its own feet.
Re:Interesting read. (Score:2, Interesting)
You make money by shrewd marketing, not by building a quality product i.e. Yugo's inception, Microsoft, Sun, etc.
But there are exceptions to that rule, build a quality product and market it "just good enough" like Cisco, and the market is yours. This idea failed with DEC/Compaq assuming the Alpha will sell itself.
IMHO, this guy has no clue ... (Score:3, Interesting)
And Linux isn't designed to run in a virtual machine; implementation decisions that make sense on PC hardware don't fit well in a virtual machine(4). This is Linux. It's designed for Intel. It's not tuned for the mainframe hardware in which it's running.
First, let's check what his "(4)" reference points to:
(4) For example: Filling all available RAM with file buffers is great in a real machine (as it speeds I/O via caching with otherwise-wasted storage), but in a virtual machine doing that is bad (as it inflates the working set of the Linux guest, which is competing for real storage with many other Linuxes-leading to paging/swapping).
Uhh, I have never seen a VM implementation that did not give a RAM limit. So this guy is basically saying that a memory leak on one of your VM's will take down the entire mainframe. Somehow I doubt IBM's mainframe R&D staff would do this
Often the difference in Intel versus mainframe applications makes porting difficult(10)
(10) Intel uses something known as little endian; a mainframe uses something different. This is significant for certain applications and makes the port difficult.
I challenge anyone out there to name any significant piece of UNIX software that doesn't have a big-endian port
Just the way he phrased that last bit about endianness convinces me that this guy doesn't have a clue what he's talking about. I can't really know for sure though, since most of the stuff he talks about is beyond me. But, based on those few things he mentions that I'm familiar with, I'd say he's a typical manager who is loosely and incorrectly paraphrasing what some Enterprise developer told him, and decided to make a marketing advantage out of it.
Read between the lines!!
Re:I disagree (Score:3, Interesting)
Hello? You obviously have never used an E15K or even a E10K with *does* domains for over 5 years now. Want some companies who use this? Ebay for one done. And they would be crazy to use Linux on it.
Re:I disagree (Score:2, Interesting)
I agree with most of what was said in the article, however, I would disagree with their statement about Linux, re: "This is Linux. It's designed for Intel.". Linux is highly adaptable, while I would agree that it doesn't scale up too well, to say that it is "designed for Intel" is a bit much. After all, Linux runs on a wider range of hardware than Solaris does.
Solaris has a very good kernel, in my opinion, superior to Linux. I'm no kernel hacker, but from a user's point of view, I have yet to run into a root exploit bug in the Solaris kernel, it has never corrupted my filesystem, and it is configurable without recompilation. The OS also contains a lot of useful stuff that you don't get in Linux, auditd for instance (Linux versions don't work on SMP systems) and disksuite is really useful. But it also lacks a lot of stuff, I usually install massive amounts of GNU software onto a Solaris install, including tcpwrappers, gnu tar, make, and openssl/ssh.
For a mid to high end server, Linux is really not a substitute for Solaris, any more than MySQL is a substitute for Oracle. At the low end Linux rocks, and any other choice is money down the drain.
Re:IBM milking mainframe monopoly (Score:1, Interesting)
You can't argue with success. It is clear as day that Sun is getting eaten up by Linux, and old rival IBM. Sun was caught with its pants down. Instead of Sun offering a comparable product, their cupboard is bare. All that is left for Sun is name calling without substance.
Re:I disagree (Score:2, Interesting)
Test Plan Omega--VM running in Basic mode on a ZZ7 (12-processor G6)--did 97,943 images.
No one, to my knowledge, has seen how far you can push it with 64-bit zLinux on a zSeries box.
Interestingly, you *can* add horizontal capacity on the fly with Linux under VM, although I don't think anyone is doing it yet. Here's how. Let's say that your bottleneck is handling incoming SMTP, and that you have some instrumented way to measure this--say, something that emits an ARM correlator when a message is received and another when it's delivered, and that this information is being fed back into something on VM that determines the average elapsed time per message has exceeded your acceptable threshold.
What you do then is clone another image and configure it in the role of mailserver (the basic IP and hostname configuration requires changing three lines in
Your old machine had Y total virtual machines, of which X were providing mail service, so the proportion of total machine time spent servicing mail is X/Y (yes, this is oversimplified; not all your machines will run at the same priority). In the new scenario, the proportion of the machine used for mail is (X+1)/(Y+1), which is larger than X/Y as long as Y>X.
I side with Sun on this one. (Score:3, Interesting)
So you are going to have to "engage" IBM Global Services to run the thing -- probaly a project manager @ $275/hour and a one or two consultants @ $200/hour.
Add to this the INSANELY expensive hardware and software maintenance charges every year and you are talking about a serious amount of cash for little benefit
When you consider the alternatives, it makes even less sense. You can buy 100 Sun E220's or 2-processor intel 4U servers for the cost of one mainframe that lets you emulate 20 Linux boxes.
Mainframes have been on the wane for the last 20 years for good reason -- they are too friggin expensive!
Re:Of Course not! (Score:2, Interesting)
Same critisms apply to Sun vs Windows or Linux (Score:2, Interesting)
Yes it harder to find technical people who know sun over windows or linux.
2.Web serving on sun box is a waste of money
Why buy one big box when many cheaper ones will do...? Save the big box for the Database server. Tux is much faster than anything from Sun.
3. Not many applications available
Most of the good sun stuff is the ported open source stuff that you have to go to ibiblio to download. Applications were mostly written on intel (SAP, Oracle) on linux and then ported to sun. Free stuff is almost all available for linux first.
4. Incompatibility across versions
The linux distros now have more in common than the big linux distros. They are more posix compliant as well. For example Sun threads and sun libraries are not portable to AIX or any other OS.
5. Propretary hardware
Yup - have a look at Suns network cards... nice but they ain't standard pci cards. 5,000 each for the nice ones. I have found the big vendors scr*w you at every turn. Memory upgrade - 50x the x86 price...
6. OK for database servers I still see the need. That is until postgresql or some other db vendor supplies fast distributed rdbms.
Actually I don't have anything against sun - they are probably one of the more "open" unix vendors. But they play the game just like everyone else.
my 2 cents worth................
Re:IBM milking mainframe monopoly (Score:2, Interesting)
IBM is milking their big win over Amdahl and Hitachi, a 30-year-plus war that took a final round of reengineering costs associated with going 64-bit (keep this in mind- could be replayed with AMD and Intel). This is fine for the Three Initial Companies that can afford this, but the future of the mainframe is in doubt, because of
*the easier entry into Unix/WiNT servers,
*the relatively slow speed of application development with traditional mainframe tools (sometimes that's a win, you don't build the pyramids with a mobile home mindset),
*the lack of z/OS personnel being churned out due to a mindshare failure/lack of mainframes at universities/tech schools,
*and new applications not being created that will bring in the new business.
So IBM needs Linux badly to grow the new businesses into the mainframes, keep the profits up with higher production runs, get those wacky penguin kids to develop new apps that sell boxes and in general keep the gravy train running even if z/OS fails due to lack of personnel/ business/ coolness whatever.
Sun, AIX, MS and Company are already nipping at the heels of the low-end part of the MVS line, so IBM drops those prices dramatically. As Moore's Law continues it's inexorable ways, IBM will be forced to slash the low-end to medium box prices to keep the TCO in line. So many companies that might eat the conversion costs to get off the expensive hardware may stay and be inadvertent beneficiaries of the Price Wars.
The power failure aspect of mainframes is priceless. If there is a power failure in the mainframe world all you have to do is wait until the lights are on, power the mainframe and associated SAN back on, and chug along while the Unix tribe fscks the night away. I'm sure journaling helps (gosh, had versions of that on mainframe DBs since the 80s), but nothing beats true bulletproof hardware.
IBM mainframes use every single ounce of processor- we can top out at 100% and everything just runs slow instead of at risk of dying. Not sure whether 390 Linux will handle it but the underlying VM sure will.
Another factor not mentioned is that mainframe software is charged by the MIPS, not site-licensed, so each move up in processor costs you far more in software costs then merely the machine and support. The Linux 390 environment doesn't have that MIPS exponential cost curve, so that aspect looks mighty tasty to sites that want mainframe bulletproof hardware and relatively cheap Linux software implementation.
Finally, we have a person who straddles the world of UNIX and MVS (he HAS to- MVS requires Unix Services to function in a non-SNA world). Our Unix manager was a former mainframe field engineer, as was our current Network Services Director. I've had three OS environments shot out from under me, and I haven't been relegated to the scrap heap. My career goal is to be able to sysadmin in both JCL and Perl, MVS and Linux.
Have a little flexibility guys, it will stand you in good stead over the decades.