Is zLinux really such a great choice for the mainframe?

[Anon Anon]

IBM's embrace of Linux on the mainframe is typically hailed as one of the factors that gave new luster to a platform that, a decade or so ago, faced a somewhat uncertain certain future. But a recent article in z/Journal takes a different view. In fact, the author says the choice between running a mainframe workload on zLinux vs. z/OS is causing a "real headache" for mainframe shops.Not long after I read this piece, I got a note from community member vatsalraicha, who said he'd like to see a discussion of how many people are using zLinux
and z/VM in production and any observations folks might like to share about their experiences.So, where do you come down on the issue? Is z/OS the only operating system to consider, or are you sold on zLinux and z/VM?Share your thoughts and experiences in the comments below, and don't forget to take the poll. (Remember, you'll need to log in or register if you haven't alreadyin order to comment or vote.)

[Anon Anon]

causing a "real headache" I think most of the pain here is in institutionalizing zLinux. Workloads will vary, and as long as you pick the right platform for your workload you will be ok. Most of the processing we do is best suited to zLinux, and so far we have no complaints.

[Anon Anon]

Great insight, ndefreitas. Would you share your thought process about what kinds of workloads are better suited to zLinux vs. z/OS and vice versa? Input from the group is welcome, of course.

[Anon Anon]

Like has been said, it really depends on what you are working to do.  As someone else has pointed out, if you have 100 systems and the monitoring of them takes 1% of a cpu on each system, when you have 100 of them - that is using 100% of a CPU just for monitoring.  Not exactly how I want to spend my CPU budget.  - - So you need to be smarter on how you monitor the systems.Some people turn off monitoring in total which is throwing the baby out with the bathwater so to speak.  Then when you have issues, you have nothing in history to go against and you have to start from ground zero to find out what is normal. Putting something like a monitoring system on there does not make sense either, since that is a constant polling system.  Like putting the Nastel system or CA Wily MOM on there.  That does work even when the rest of the systems are doing no work at all. it is great for a test bed where you have systems that come and go with workloads, but we are doing that on VMware and that seems like a cheap soltuion as well.  Loading up a VMware cluster and then letting that spread the work around is far cheaper for us than IBMs solution set.Once you have more than a couple of special engines, then it starts to make sense to run more on the zVM and zLinux systems, but if you have just one or two engines, it is hard to make a case to do much on them other than science tests. Just my view point.Henry

[Anon Anon]

Good observations, Henry. The results of our highly unscientific poll seem to back up your assertion that zLinux isn't widely used beyond, as you put it, science tests. Well over 60 percent of respondents are just using z/OS; a little more than a quarter are using some combination of z/OS, zLinux, and z/VM.And yet, there seems to be a lot of interest in zLinux. Something to watch.

[ca.portal.admin]

I think it could be a very interesting idea. But in my opinion, the entry costs to Linux on mainframes is much too high!Wolfram

[Anon Anon]

...the entry costs to Linux on mainframes is much too high! I agree that Mainframes are not cheap, however technologies like zPrime (NEON), IFLs, zIPPs, and zAAPs enable companies that already own Mainframes to defer or avoid more expensive General Processor upgrades (and associated 3rd party software license costs). zLinux & zSolaris will continue to allow these companies to consolidate their distributed server infrastructure while enjoying the low cooling/energy/space needs of the Mainframe. If you're a purely distributed shop, and your server inventory is rapidly growing it may be cheaper to consider a cloud solution (such as Amazon EC2, or the Google AppEngine) - there's a lot more that goes into this decision such as data security requirements, and the portability of you applications. IBM too will need to offer private clouds for the enterprise - but odds are that it will also run on zVM in some form of dynamically provisioned zLinux/zSolaris images.

[ca.portal.admin]

Forget about zPrime. That product has nothing to do with Linux on mainframes. What I wantet to state ist that Linux on system z ist too expensive to start with. If you go to your boss an tell him that you wnat to install linux on system z he says: " Why not, what does it cost me?" You tell him then, thait it needs at least one IFL, a z/VM for vitualizing software and some administration resoorces. All that stuff adds very fast to 100k.And that ist really too much for a system you joust want to play with or migrate some programs.WOlfram

[Anon Anon]

In light of the good debate going on here, I thought I'd take a minute to update the results of our (still highly unscientific) poll. A full 72% of respondents use z/OS, with only 17 percent saying they use a combination of z/OS, z/VM, and zLinux.Whether that's due to the expense involved or other factors is beyond the purview of this poll, but I'll leave that all of you to continue to debate here.Enjoy!Kim

[Anon Anon]

Kim,it would be an interesting number to see how many of the 17% that are using a combination of how many z engines they have in their complex?I would bet that most of the people in the 72% hax just 1 or 2 z engines and that is mainly to do science experiments with and not for real production use.Those that are in the 17% area probably have more than 5 z engines and therefore have enough to make it worthwhile to actually do other work with.   Once you have enough, then it is worthwile to use them for more than a simple test here and there. Henry

[Anon Anon]

Hi Henrys,That's actually a great question to put to ndefreitas, who, so far, is the only voice here who seems to be using zLinux in production. ndefreitas (or any other members who are using zLinux in a production setting) would you be willing to share details of how many z engines are in use in your complex?Thanks,Kim

[Anon Anon]

I'm a little late to this blog party, but thought I would chime in with my 2 cents.  I am not sure about the scientific approach of your poll. Another poll mentioned by the Mainframe 2.0 showed significant production usage and, more importantly, significant projected growth in z/Linux over coming 2 years. In my last job, I asked the CIO if anyone had considered z/OS-DB2 and z/Linux for SAP instead of the 21 flavors that were then being implemented by the various business units. He stared at me blankly for a minute then said "That might not be such a crazy idea".  That was in 2005. That company now has twin fully loaded z/10's, just for running SAP corporate wide. A detailed outside analysis showed that the mainframe implementation would save the company 50% of the projected cost of implementing a Solaris/Oracle solution at the same scale. When we began looking at it, we talked to Wells Fargo who had migrated a significant portion of their distributed UNIX and Windows environment to z/Linux. I believe they presented at Share (2-3 years ago).Also, someone mentioned the entry point for mainframe and z/Linux. I think they said the entry point for 100 servers would be $100k.  If you only have and projected 100 servers, okay, the mainframe is probably not worth it. But if you want to start at 100, with an eventual of 500+, the a simple ROI will take you to mainframe pretty quickly. Plus, IBM just announced a z/Linux ONLY mainframe. IMHO, IBM is setting up the mainframe with z/VM and z/Linux, z/Solaris, and z/Windows, to become the platform of choice in the enterprise-class Cloud markets. The mainframe z/VM-z/Linux-z/Solaris-z/Windows package is almost a Cloud solution by definition. Will large enterprises go to what they know, mainframes, or to something that has outages every 2-3 months, Google, Amazon, Azure, Rackspace, etc. Anyway, next year's z/11 announcement should be very interesting, especially with the rumored x/86 linkages.I believe the question for 2010 will become, should an enterprise with a mainframe now get rid of ALL other servers?We are in for a Disney E-ride....

[Anon Anon]

Hi KLCameron,You're not late at all--this is an ongoing discussion and one, as you note, that will likely get even more interesting in 2010. I, too, have seen the kinds of studies you mention that point to increasing interest in Linux on System z (and I'm the first to point out the unscientific nature of our poll). What do you make of the new Linux-based mainframe from IBM? I started a new thread about that over here, but I certainly won't object if you want to answer on this thread.Thanks for speaking up!Kim

[Anon Anon]

To answer your question above effectively it really depends on what the initial goals are in moving to Linux on the mainframe. Certain workloads and applications are better candidates to move to Linux. The application areas we have seen customers move most to Linux on System z include ; Web Serving and Web Application Serving data serving and systems developmentA recent customer that saw benefits of moving to Linux on System z is Blue Cross Blue Shield of Minnesota - http://tinyurl.com/y8hvvb7 BCBS achieved a significant TCO reduction with virtualized SUSE Linux Enterprise Server for System z. For other proof points on customers that made the switch to Linux on System z you can visit: http://bit.ly/6EcdAf

[Anon Anon]

Hi Kimberly,Thanks for pointing that use case out and for the link to more. I read an interesting blog yesterday from Marcel Hartog. He argues that, while mainframers understand that the mainframe is perfectly suited for use in a cloud, the only way those outside the mainframe world will understand that is if mainframers demonstrate it to them. Naturally, his words are better than mine--you can read the whole post here, but here's an excerpt: Everybody wants "cloud". And at the time we are telling other
mainframers how good the mainframe is, the "Blade people" are building
clouds. They are demonstrating to Architects and IT Management how easy
it is to run 12 images on 2 or 3 physical boxes and they demonstrate it
as a mini-cloud. More a puff of smoke, but the architects and your IT
Management sees "Cloud" in action. And they like what they see. So the
architects start making plans to build the next generation IT
Infrastructure. And in these plans, I do not see the "Z" word...
Believe me, I see it happen around me..........

I have learned an important lesson in the past: If you want to
convince people about something, you start by talking about it, but you
very quickly switch to actually DEMONSTRATING the point you are trying
to get across. So let's stop talking. Fire up some Linux and
demonstrate that you can run the same stuff as "they" can. More
flexible, more efficient, more powerful and better managed. Demonstrate
that you CAN get 20 year old virtualization techniques to work for your
company. Demonstrate value... The tools to manage the virtualized
environment are available, so that should not stop you...
Kim

[Anon Anon]

Thanks for sharing this blog. I'm new to the blogosphere and hadn't seen this one yet.  I also read one of the articles referenced by Marcel  "IT Management: Top-10 Reasons to Look at Mainframe Cloud Computing". While there are some earlier adopters like Tranzap as referenced in "Cloud Computing for the Mainframe: Will It Ever Cross the Chasm?"What does this group think- will cloud computing on the mainframe cross the chasm?

[Anon Anon]

Thanks, Kimberly, for the pointer to that article and suggesting a new area for discussion. I've started a new thread where we can discuss what the adoption of cloud computing means to the future of the mainframe. Check it out (and add your thoughts) here.

[Anon Anon]

This is a strong area of interest for me. We are looking at Linux on z as a server consolidation play. The only problem is, that commodity servers are so much less expensive than the mainframe in terms of acquisition cost. It is difficult for management to look beyond that to what the fully loaded environment costs to run and maintain. In a z/OS to zLinux comparison, Linux on z is the clear winner. Compared to commodity servers running Linux however, not so much.The Linux on z Solution Edition seems to be setting the playing field  a lot more level. However, many challenges remain. I'd be interested to know if any companies successfully moved all or part of  their commodity server infrastructure running Linux to Linux on z - and if so, whether overall cost was a factor.

[Anon Anon]

Hi StevenWe have a few examples of customers who have indeed moved a large portion if not all of their environments over from Linux in a distributed environment to Linux on System z. In some customers llike Nationwide and Dundee Council customers do indeed chose to keep some workloads on x86 platforms while leveraging the benefits of Linux on System z for other workloads. While costs are often a key driver for making the move from a distributed environment to Linux on System z other factors play a role in deciding to move workloads to Linux on System z. These include: energy consumption, reducing floor space, managing more virtual servers with fewer people and deploying new servers and applications faster.Atos Origin recently deployed an IBM Eneterprise Linux Server with SUSE Linux Enterpise Server and had the following to say ""With a traditional distributed approach – even using the latest virtualization-enabled x86 processors – you end up with a lot of unreliable boxes to manage. With the ELS, we can run hundreds of environments within a single physical footprint, and easily deliver the 24x7 availability that our customers demand, says Colin Clews, Technologies Manager"Also if you haven't seen IBM's TCO tool for System z- take a look- this may help you as you evaluate cost benefits of Linux on System z.http://www-03.ibm.com/systems/migratetoibm/systems/z/tco.htmlIf you want to discuss more off-line please feel free to let me know.

[Anon Anon]

Hi Steven,Thanks for raising these issues. I'm curious whether there's any discussion of looking past acquisition costs to TCO over a period of time? Kimberly gives one link below for exploring TCO (thanks, Kimberly!). I think I've also seen some good discussions of evaluating the real costs of distributed servers vs. mainframes (and vice versa) in Mainframe Executive (or a related pub). I'll see if I can scare up some links to share.Keep us posted on the decision making process.Best,Kim

[Anon Anon]

Here is a new article from Mainframe Executive that may be helpful around some of the decisions regarding TCO and the mainframe.  TCO: Specialty Engines and Labor Savings Bring New Perspectives to Mainframe Total Cost of Ownership Additional artcile update from CIO. It's an older one and this community may have already seen but might point out some good advatages, such as the section on virtualization to consider when evaluation TCO: http://bit.ly/9aaPoWEnjoy!