Slashdot Mirror
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?"
Besides, Sun will attack IBM at any chance it gets.
And the men who hold high places must be the ones who start
To mold a new reality... closer to the heart
Like the man said, this sounds like the normal responce from sun, I'm sure Microsoft should have simular arguments for why MS is better. Unless we get a complete third party to analysis with no aligence to any OS. And maybe on that day pigs will fly.
http://www.funwithpenguins.com
Microsoft says the same thing. Does this make it true? No
If anything, this is a really good sign for the ever maturing linux operating system. Of Course sun would want to move people away from an open source, free operating system, over to their 'paid for' one. And if they can't do that by simply saing "don't use linux, use solaris", it makes a lot of sense for their marketers to simply say "don't use linux, its bad... and scary". It still cuts out a potential threat to them.
I figure if IBM says that IBM is ready for linux, i will trust that a lot more than solaris saying IMB isn't ready for linux.
Not that i have anything agasint sun, or solaris.. i respect sun and what its doen, and been through.. i just question the reasoning for this 'article'.
I'm not sure what to think about the fact that Sun has a "Chief Competitive Officer." Please tell me that there's more to the guy's job than spreading FUD about the competition.
This article is misrepresented as bashing Linux. It doesn't say that Linux isn't up to the job of running on a mainframe as much as it says that many of the benefits Linux offers are lost when running it on such a system -- basically bashing IBM's solution, not Linux.
Finding mainframe staffing is an obstacle in many organizations(6); combining mainframe and Linux staffing further complicates the matter. Running multiple Linux images still requires administration that needs to grow with the number of images being run.
This statement applies no matter what operating system you choose, you still have to find people who know the hardware. And as with all VM systems, you have to actively administrate each image. This statement is Linux agnostic.
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.
Again.. Linux isn't repsonsible for the machine not being able to dynamically allocate resources to over-utilized images, it's a hardware/underlying OS issue.
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
Duh. It's a different architecture.
So, SUN isn't really bashing Linux, they're bashing their competitor, IBM. No real news here. SUN is very careful not to say "Linux sucks", because they have Linux offerings, they're just saying that customers should buy the SUN/Solaris solution for their high-end systems, not the IBM/Linux solution. I'm sure we'll see something from IBM soon.
--XaXXon
Next thing to do would be to ask someone that recently switched to linux on the mainframe, like ebay... hope one of the links below still works...
http://www.cio.com/archive/010101_et_content.htm l
http://abcnews.go.com/sections/tech/DailyNews/ib ml inux000517.html
http://www.zdnetindia.com/biztech/resources/ebusin ess/ecommerce/stories/45234.html
Only 'flamers' flame!
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. Heck, even poorly written native compiled code generally has advantages over such a set up.
There are however, notable exceptions, given the nature of mainframe processors, if all of your apps are written unoptimized for such a system, then you would want to unify them in a familiar abstraction, given a close enough match, this makes Linux a natural choice. Of course, why would you buy an expensive mainframe and not optimize for it?
To the naysayers slamming Sun as merely trying to boost SunOS, well, yeah, they are, but lets look at the situation.
1) Sun still has SunFire servers, which are QUITE powerful.
2) Solaris is no longer competing with HP-UX, since HP-UX is no more. Sun sells windows and linux based solutions. In other words, Sun has no reason to just blindly nay-say against Linux. As far as exploiting Linux for being a hot technology, well, they're doing that too. That's business for you, you gotta do what you gotta do.
In otherwords, the z800 isn't exactly slaughtering Sun's business, but you gotta have whitepapers to back up your statements when you're bidding to large customers. Saying "just cuz" isn't good enough. Sun's scoring one for the people who want to buy their products. It's not "slamming linux."
sounds like they have more of a problem with IBM than with linux.
-... ---
He basically has 1 valid point. Linux under VM is not as well behaved as it should be, for exactly the reason mentioned. Coming from the PC world where all hardware is real, it treats all memory as if there isn't anything else better to do that use it all for file buffers. Under VM, a better plan is to check to see if there is any memory pressure being exerted on the machine before using your entire memory allocation for buffering.
They're working on that.
As for the rest, it's mostly FUD. The endian-ness is not an issue for 95% (wild ass guess) of apps that I have seen. Maybe except for DB2. You have to plan your maximum capacity in a discrete server farm just like you do in a virtual one. You also get capacity upgrade on demand with a phone call with the IBM hardware. They dont even have to send out a CE to do anything. Let's see SUN do that.
You wouldn't want to use it as a compute farm, but as a database server or news server or something which is usually I/O bound. They ain't exactly ferraris, more like 18 wheeler big rigs.
You think emacs is evil?! You've never used VM's XEDIT have you?!! That's evil, baby!
You are talking about a different deployment than the one that is being attacked in the Sun article. What the latter is discussing are multiple images of Linux being hosted on top of a VM.
There is no reason why you should have been doing that in your case: you should have dispensed with the VM layer and just used Linux "native".
Basically the article is Sun bashing (perhaps righlty or wrongly, I don't know) the concept of "server farm in a box", which is completely different from your task!
As far as I know, Linux wasn't really designed to be a mainframe anyway... so whats the big deal really? Linux is a web server/ server system, great for apache and the likes, stable and a life saver for small companies. So Sun has the better equipment and software for mainframe. IF linux had been designed to do that then there would be a real contraversy in the situation anyway. Its an interesting article with lots of good points, but its like comparing tomatoes and oranges.
-- RJ
Well then, you should buy a starfire, because you
can keep it running, slap in new processors, new
memory, and then suck them into a running
partition.
It seems that most of the criticisms of Shahin
Kahn's article are based on ignorance. It's a
fair assessment of the liabilities of using
mainframe hardware for typical modern web service
applications. IBM tried to save the mainframe
from declining market share in a very ingenious
way, and Linux and IBM have benefited from it,
but that doesn't mean that it is competetive
with Sun's hardware offerings for the same
application environments.
Not all of Kahn's objections to VMs are valid,
however. The robustness arguments are good, but
the performance ones are short-sighted. While
s/390 Linux may not be tuned today, you can be
confidently assured that it will be soon -- even
if IBM has to fork the kernel to do it.
-I like my women like I like my tea: green-
With regard to the linked benchmarks, somehow I bet the benchmarkers at Tom's didn't compile parallel. Of course there is not going to be a speed improvement if you don't run a parallel make!
As someone who builds a small embedded Linux system from scratch (including gcc and glibc), a dual processor system is VERY nice. It cuts down the compile time by at least 30-40%. make -j2 is your friend with two processors.
They are both Unix(-like), both solve the same problems but Linux does it cheaper and allows you to look ath the source code.
Actually, the price of Solaris is not really that expensive unless you are using a system with very many CPUs. A single user copy if you download it is pretty much free. It is the hardware needed to run it on that is expensive. Also, the solaris source code is available.
I doubt Sun really cares that badly about the success of Solaris so much as they care about the failure if IBM. As far as I can tell, this article is mostly pushing the fact that a cluster of low end Sun boxes running Linux will be better than an IBM mainframe running Linux.
The clash of honour calls, to stand when others fall.
Friend or not, in the long run Sun will need Linux for Sun just to survive. So the question isn't is Sun sincere in their Linux efforts nor does this attack on IBM signal new antagonism between Sun's interests and Linux, but instead do we even need to notice Sun when it says things like this?
The answer is no, of course not.
We neither plead with Sun to get onboard with linux nor get angry and bothered by this kind of attack. It's 100% irrelevant.
It's up to Sun Microsystems to come to Linux and save itself; nothing we do or say can positively influence their intransigence/willingness, so we're best off ignoring them completely no matter what they say.
Think how absurd it is for Linux people to get upset or harbor paranoia about Sun!
Without doing anything hostile to target or hinder Sun, or Sparc or Solaris, Linux has compelled Sun to include linux compatibility runtime, to offer an official JDK for Linux, and to acquire a Linux on X86 hardware business and now, to expand it into the midrange of server offerings.
Linux is an irresistible force and, no matter what some dinosaurs over there may think, SUN is not an immovable object, it will orbit Linux or it will fall into Linux and burn up. Whether they survive the journey or not they will come. They are coming over even now.
This piece was so full of FUD that I could scarcely believe it.
Then why did IBM have to port the TCPIP stack from the VM world into MVS, if VM is so far behind?
Linux is "designed for Intel"??? What about M68K, PPC, ARM, Alpha, SPARC and others? (See the Debian ports page for a more complete list.)
Yes, MVS (z/OS) is rock solid reliable. But the machines don't bust either. CPU recovery has been an integral part of the architecture for almost 30 years. If a processor breaks, another takes over with no application effect, or a spare is assigned. Someone on the ibm-main list today mentioned that the processors are themselves duplicated on chip, with comparison logic to ensure that both sides are computing the same thing. Does Intel even parity-check their processors?
Small? Install a copy of SuSE SLES in a S/390 LPAR (logical partition, a hardware implementation of VM that is delivered on EVERY S/390... no z/VM necessary) and see how much software was delivered with it. You wanted OpenSSH and OpenSSL, though SuSE didn't deliver it? Go to the web, download it, and do configure, make, make install. The big problem with application portability is the proprietary vendors that ship binaries only.
What an amazing assertion. Wish Khan had provided a reference.
Merde. Why run it on a closed proprietary SPARCstation? Or a closed proprietary Mac?
Khan makes a couple of decent points, particularly regarding z/VM skills. But the hyperbole is way out there, and it's hard to take him seriously.
You've got to be kidding me.
The economy and most business models are not a 100% research and development, not-quite-stable environment.
Don't tell a large business "Well, it will get better the more people who use it". They'll spit in your face. They need to know what works, and what works now, and what will continue to work in the future.
Right now, Solaris works. Linux-bigots will sit and say "Well Solaris doesnt provide useful GNU utilities and is a boar when it comes to performance!" Well, yes it is, but it's been around forever, and when Sun says they can make it work, they will MAKE it work. You can't sit around and play with something for awhile in a 100% production environment, and rely on tools which have a sketchy (in a business-model sense) support base. It just cant, and wont, happen.
Just my $0.02.
Maybe Sun is jealous because they *still* have no real enterprise level HW/SW combination that can do real clustering and partitioning. Just compare the stuff Sun can offer with their competitors (HP and IBM). Sun-cluster has downtime (at least with certain applications) up to several minutes and that is not what I call enterprise level operating environment. With IBM HW you get partitioning and abstraction that will you replace every single component in the system with zero downtime. And what comes to performance, I'd recommend cheap Intel boxen to all customers instead of Sun, they're basically identical when it comes to capabilities and Intel just runs Java much better (just try WebSphere, ATG Dynamo or even Tomcat).
I know Solaris is something like the standard Unix in the corporate world and I hate AIX and HP/UX, but the HW just is so much better. Now, with the rise of Linux, you get the superior HW with decent Unix-like OS.
You can't really compare a zseries on computation power. This machine is extremely competitive on I/O (serving static pages, doing database queries, read/writes, etc.), but is not very strong on computation. If you want to do computations with IBM equipment, use their pSeries line, which is risc based, or their intel line - the xSeries.
For your specific application, you - and your vendors - should have spared themselves the trouble. All the tweaking in the world would not make a mainframe with 20 CPU's that costs some $2 millions comparable to a cluster of intel boards of a similar cost. That's is why these machines are not used for modeling and rendering, but for payroll, inventory, etc.
...there are other factors than performance, indulge me for a moment:
I work at a big corporation which relied on IBM mainframes for its whole business for almost 30 years until the PC and the high-end Unix servers shook up the landscape for good. I'm from the PC (IT) camp, which has been separate from the Big Iron (DP) guys in the organization since the early days.
DP, once very powerful, has lost a great deal of influence in the 90s, although they still run most of the mission-critical stuff, and the main reason for this were the high-end Unix servers, most of them Sun boxen running Oracle. Believe me, there's no love lost between those two fractions in our company.
Our mainframe guys see Linux as an opportunity to get better integration with the IT world, which was abysmal until now (3270 terminal windows, IMS/DB, TSO/ISPF and such horrors) and to better position themselves against the Sun/Oracle camp which is after their budgets and their butts. Today, we have Linux happily running on our mainframes (still in an experimental phase, not in production), serving up http and Samba shares without a hiccup.
If we're talking about bringing Linux into the large corporations, the crucial influence of IBM cannot be overestimated. We were a died-in-the wool IBM shop (S/390, Token Ring, 3270PC, OS/2, S/36, AS/400, the whole enchilada) and successfully trusted our business to IBM for 30 years (paid through our nose for it, too, I might add). IBM has lots of credibility and trust, so if they say Linux is cool, our CTO listens. Microsoft, on the other hand, is viewed with some "new kid on the block" suspicion. Our management doesn't like downtime and security breaches, and the memory of the ILOVEYOU aftermath is still very vivid, for example. Plus, we migrated to NT4 late (about 28'000 systems, ended September 99) and now Microsoft is practically forcing us into another expensive upgrade cycle sooner than we wanted and with IT budgets cut short on account of the less-than-stellar economy because NT4 support is withdrawn in 2003.
We thus have the following situation: IT and DP are up against the Unix enterprise server guys, all this with the backing of IBM. The astronomically high cost of Sun/Oracle solutions is being questioned more and more, and technologically viable low-end solutions (x86 multiprocessor servers, Linux) begin to rattle the foundations from below.
I don't want to make bold predictions here, but if I were Sun, I'd be worried. To me, it looks like interesting times are ahead.
"There are already a million monkeys on a million typewriters, and Usenet is NOTHING like Shakespeare." - Blair Houghton
Linux on the mainframe can't respond to the workload demands of Web serving with high utilization--something IBM touted at the time of its z800 announcement. Horizontally scaled Linux farms are designed to handle unpredictable demand with above average peak loads. As demand rises, a load balancer distributes the traffic evenly across servers, which increases utilization. Because design capacity needs to handle peak demand, server farms often have a low utilization.
If you have a VM system with two virtual machines, and one of them is nearly idle, and the other virtual machine is very busy, VM will automatically take resources away from the less busy machine and devote it to the more busy machine.
This means that you don't need load-balancing software. VM is the load-balancing software.
Given the relatively low cost of hardware, some organizations find this trade-off acceptable to ensure appropriate service levels. Contrary to what many believe, consolidating a Linux farm into multiple images on a mainframe would not change the demand pattern. Although z/VM can start and stop Linux images, it cannot dynamically add resources to match demand.
Of course it can! The VM kernel will parcel out memory and CPU on demand.
As a result, a mainframe would need to size for peak demand just as the Linux farm would;
All computer systems need to size for peak demand. The difference is that with a mainframe, you can size one machine for the peak demand of the busiest of a large number of virtual machines, and get rid of the overhead caused by the load-balancing software, because you don't need it anymore.
high utilization is a myth.
VM systems can utilize 90-95% of the native computer resources. The overhead on a VM system is very, very small.
Naturally this article is going to be met with some skepticism because it appears to be a self-serving marketing piece from Sun. This is unfortunate because the author makes several good points. I think it is important to note that this is more of an attack on mainframes and VM architecture than it is against Linux. It really does not make much sense to run 10, 100 or 10,000 copies of Linux on one super-duper computer. Sure it's neat, but we need to remember that computers are supposed to do useful things. What a collosal waste of cpu and memory to have 10,000 operating systems when the right one (yes, 1) would do the job just fine. Add to that the inherent performance hit from running on a "virtual machine" and it makes even less sense. I actually tend to think that IBM's use of Linux is more self-serving than Sun's attack. It's just an attempt to sell more expensive proprietary hardware by capitalizing on the intellectual generosity of others.
-
Linux on the mainframe is actually hosted by
another proprietary operating system, z/VM.
-
The optimized operating system for IBM
mainframes is z/OS, formerly called MVS(2).
Compared to z/OS, z/VM is a niche operating
system with virtual machine (VM) support for
new hardware features added late or often not
at all(3).
- Linux on the mainframe is complicated; this
isn't Linux running on a two-way Intel server. Despite
IBM's claims of easy management(5), customers still
need a special machine room and specially
trained staff for both z/VM and Linux.
-
Finding mainframe staffing is an obstacle
in many organizations(6); combining mainframe
and Linux staffing further complicates the matter.
-
Running multiple Linux images still requires
administration that needs to grow with the
number of images being run.
-
...there is little incremental RAS benefit.
Although IBM claims "zSeries servers inherit
the legendary IBM S/390 strengths in the
areas of fault avoidance and tolerance,
recovery from failures, and concurrent
maintenance and repair for "always-on" availability"(7).
We don't believe this to be true for zSeries
servers running Linux. The "legendary" IBM
S/390 strengths IBM references are the result of
decades of development work on IBM's flagship
mainframe operating system, known today as
z/OS. The fault recovery features of z/OS are
not found in Linux. z/VM does have some fault
recovery features, but it is not nearly as
resilient as z/OS. For example, z/VM cannot
take advantage of Parallel Sysplex clustering,
and VM hypervisor is an added single point of
failure(8).
-
Consolidation without application availability just can't happen...
-
Often the difference in Intel versus mainframe
applications makes porting difficult(10).
-
The economics just don't work. IBM claims it is
financially justifiable to consolidate as few as 20
Linux servers on a z800(11). With an estimated
starting price of $400,000 for a z800(12) with a
single CPU engine enabled, that claim seems
exaggerated compared to Linux servers that hover in
the $1,000 to $2,000 range. Sun's rack-optimized
1U form factor servers start at list prices under
$1,000.
-
Customers such as Nielsen Media Research, A.B. Watley,
Cognigen Corp, and Littlewoods have chosen Sun and
realized up to $1.5 million in annual savings
-
[out of order]
This is Linux. It's designed for Intel. It's not
tuned for the mainframe hardware in which it's running.
Don't get me wrong...I love Sun, have an Ultra 5. But a lot of this is obfuscation. There is a lot of value in Linux on the mainframe for the huge organizations which are, in the main, already IBM customers.Not necessarily, you can boot Linux as a stand- alone operating system. It's just more convenient to develop and implement under VM in many shops.
There is a grain of truth in this. Some printers, for example, were supported late in VM. I think the same is true of some tape libraries. However, it's a real strong niche, and usually this is not an issue.
Linux on anything can be complicated. If it follows the historical path of IBM products, it will get both (a) more complicated as IBM adds options and (b) simpler to administer as they clean up the packaging. But the real problem here is the specially trained staff...you need a trained staff no matter where you are running Linux...or any operating system. The same is true of Solaris or Windows servers.
This is a good point. Mainframe skills are becoming rarer, older, and grayer. But...once you have the VM systems programmer, the rest of the Linux staff can all be new hires trained for IBM Linux. However, finding good Solaris skills can also be a challenge.
No. The strength of IBM is scalability. The strength of IBM customers is organization and procedures - you might see it as bureaucracy, but it has gotten the job done for two generations. And, if I am running thousands of Linux virtual machines, I will be able to automate the management of these with existing MVS mainframe tools.
There is a lot of machine check recovery processing in MVS. However, over the past decade a lot of that has moved into the hardware itself, and into the Hardware Management Console (HMC). So, this was certainly true in 1990, less so today. Ditto the hypervisor comment IMHO.
Unless you are writing the applications yourself, as many potential IBM Linux customers are.
Little-endian vs. big-endian. Hits Solaris too.
Just looking at hardware and/or software costs isn't the right analysis for most of these customers. You also have to factor in the cost of support staff; comparing z/VM and Linux to a corresponding Windows NT server farm becomes more competitive as the size grows. And factoring in the cost of downtime, which is the real driver for the very large companies who benefit the most from this combination, usually makes the hardware costs insignificant.
Neilsen is a household name, haven't heard of the others, but...these are hardly huge DP shops. The real benefit of z/VM and Linux is going to be in the Fortune 100 manufacturers, insurance companies, banks, and stock brokerages whose IT budgets approximate $1.5 million per week. Or day.
This is Linux. It's designed to be open. It's tunable for any hardware.
Solaris may be more scalable, but more stable? It depends very much on the hardware.
To pull an example out of the air, I don't think I've ever seen a stable Solaris on the Sparc Ultra 5 (their 'cheap' IDE-based workstation.) I've also witnessed some really nasty wedging with LDAP authentication and panics on Sun Ultrasparc machines fighting it out with Sun-branded RAID arrays under load.
Maybe the latest version of Solaris is completely unlike its predecessors, but it seems a little unlikely.
Some versions of the Linux kernel, together with XFree, GNU software and other tools, have been exceptionally stable on certain combinations of x86-based PC hardware. Given that Sun control their own hardware, it seems unfair to criticise Linux's stability when compared with Solaris.
their web server times out on me. Why?
Because Sun's firewall is broken and drops TCP connections using Explicit Congestion Notification (ECN).
How stupid can a single company be? I bet that article goes to great lengths to say how well suited Sun is to provide scalable web servers. And they can't even get their own web server configured properly!
What a buch of losers. ECN is, by the way, an official internet standard (RFC3168), which happens to be implemented by Linux.