Linux on System z

Black Friday Special – World’s best cloud hardware platform for free

(or close to free)

What if I told you that there was a platform for your on-premise, private cloud infrastructure that gave you greater than five 9s hardware uptime, was the ultimate in performance (especially for IO), was the gold standard in security and would integrate seamlessly with your other core systems? What if I also told you that this solution would not add ANY additional requirements for data center floor space or power consumption and would reduce your license fees? What if I told you that you could get access to it for free or close to free?

Well it exists and the only proviso is that you need to own an IBM Series z mainframe. Don’t laugh; plenty of companies own them and they might be missing out on a huge opportunity to leverage the mainframe’s super resilient architecture for other work.

The secret? Run your private cloud infrastructure on IFLs on your mainframe. (If you don’t know what an IFL is then take a quick look at the “What’s an IFL and why should I care?” entry and then continue.)

So here’s the picture. You have your mainframe running your business-critical, number-crunching workload already. It has zero downtime, high security and phenomenal performance. You take your cloud infrastructure, (currently running on those distributed servers), and you move it over to virtual Linux servers running on the IFLs in your mainframe. Now that you have your mainframe and cloud payloads running on the same physical device consider the ability to inherit all of the robust features of a mainframe, the ability to more easily orchestrate workloads between the cloud and mainframe and then think about the improvements in IO between the two.

Take it from someone who just spent two years working on enterprise-strength, commercial cloud deployments. Any platform that you can inherit high availability and security from is worth its weight in gold.

What’s not to love? Why aren’t more people doing it? Why don’t companies buy mainframe systems just to run their cloud deployments?

Here’s my list of probable reasons:

• The assumption that the mainframe is going to be more expensive or is a dying breed.
  • Obviously this depends on your deployment but I’ll be working on some ROI comparisons to bring some reality to this. Stay tuned.
  • The mainframe market continues to grow and I’m betting my career that with the growth of big data and the cloud it will only continue to grow.
• Lack of understanding that the mainframe is an ideal platform for the cloud, (I’ll be working on alleviating this in my blog for a while)
  • You need to understand the challenges of deploying real-life, enterprise-strength and business-critical systems to understand why the mainframes architecture solves so many issues. Most people have not deployed systems that are described with three hyphenated, compound words so they don’t get this.  When they do they’ll realize how many issues can be removed by using the right platform.
  • More importantly, you need to know that mainframes are capable of running systems that are more often associated with the distributed world. Two months ago I didn’t know that you could but I’m sold on the idea and I’m convinced that companies will start to buy mainframes with just IFLs for hosting cloud systems soon.
• Internal divides
  • The mainframe team is often very separate from the distributed team. Today most organizations leave the cloud deployments to the distributed side of the IT department and that team typically don’t really know much about the mainframe’s capabilities.
  • With the right tools the management of a cloud system on a mainframe can be done by the same team that manages it today. Let’s face it, most cloud administrators are not actually involved in the day-to-day management of the hardware anyway.

So, there it is. I am going to write a lot more about this topic. I will focus on specifics about the scale and cost of running cloud (or just general Linux) systems on an IFL over the coming months.

What’s an IFL and why should I care?

An overview of “Integrated Facilities for Linux” aka IFLs

I am going to talk about IFLs a lot over the next series of articles so here’s a reference article to explain at a high level what IFLs are.

All computers have processors; most have different types of processors inside the box each optimized to do a specific task. For example, your iPad has a processor for running the actual Angry Birds game and a separate video processor for telling the screen what to display. In theory you can use a single processor for both but you optimize performance greatly if you offload some specific tasks onto a separate independent processor.

In an IBM Series z mainframe there are four main types of processor available: CPs, zIIPs, zAAPs and IFLs. They are all the same physically but they are used for different things and priced differently.

• CPs (Central Processors) will do pretty much anything you want and are the number crunching, MIPS digesting beasts that we expect in a mainframe. They represent the expensive part of owning a mainframe but they do most of the heavy lifting too.
•  zIIPs and zAAPs are just as powerful but are limited to be able to execute only specific tasks: typically DB2 operations and Java/XML processing respectively. These are not included in the MIPS pricing equation and have a significant lower cost of ownership.
• Integrated Facility for Linux (IFLs) are also the same powerful processor hardware as the CPs but they are also limited by microcode to only run…you’ve guessed it – Linux. They are exceptionally inexpensive considering their potential processing potential and do not contribute to the total MIPS in a system’s pricing.

In a nutshell - you can run Linux (specifically Linux for Systemz) on IFLs on your mainframe. Typically you will run these instances of Linux on top of a System z OS called z/VM which allows you to virtualize the Linux instances and run multiple copies on a single IFL.

Note I am not referring to the zEnterprise BladeCenter Extension (zBX) which allows you to run Windows, standard x86 flavors of Linux, etc. No doubt I will later though.

What’s Next? Cloud Applications…on a mainframe.

I have taken a new position with CA Technologies to manage products that let you design, build and manage cloud applications on a mainframe. Yes, you read that right, “on a mainframe”.

I’m not going to comment on why I took a role associated with the cloud; it is no secret that I am convinced that cloud delivery of enterprise solutions is the future. In all seriousness, if we can deliver ECMaaS then we can deliver almost any conceivable solutions al la “aaS”!  According to IDC^[1] “80% of net new commercial apps will be developed for cloud in 2012”.

The question you might have is, “why would one want to host cloud applications on a mainframe?”

The first thing to understand is that the mainframe is not the equivalent of the dinosaur in the Cretaceous Period. The mainframe market is huge and growing healthily. According to IBM “More than 80 percent of corporate data resides on mainframes” and common estimates suggest that 85% of all business transactions are processed by mainframes today.

Once you start to think about using a mainframe you'll probably see some synergies with what you expect from the underlying cloud infrastructure.

• Performance - the hallmark of a mainframe system, not just raw processing power but more importantly, very high performance I/O.
• Resilience - High Availability is a real challenge with distributed systems. In order to get four or five 9s uptime in a distributed environment you typically need to have massive duplication of hardware, sophisticated monitoring, automated failover and redundant infrastructure. However mainframes are inherently more resilient to hardware and software problems, (the z in zSeries stands for “zero downtime”).
• Efficient use of available capacity - I appreciate that this might be the tail wagging the dog but if you have a mainframe and have spare capacity then being able to move cloud workloads on and off the mainframe as capacity allows can be very cost effective.
• Virtualization - this is a key feature of almost cloud solutions and the mainframe was the birthplace of this technology. Typically as mainframe capacity grew people only needed a "slice" of the system and so hardware, and later software virtualization was created.
• Security - again, this is bread and butter to mainframe systems which were designed originally to handle massive amounts of transactional data and protect it adequately.
• An OS - your cloud is going to need a "normal" operating system, it's not likely to run natively in z/OS. Conveniently the IBM System z hosts Linux using z/VM, in a logical partition or even as a completely native operating system.

One aspect that you might not have considered though is orchestration. Consider that many systems that you’d want to run in the cloud need to interoperate with a mainframe system. Let's assume that you have an online banking system running in the cloud. If your bank is successful then you'll have a mainframe somewhere at the backend doing the raw number crunching for you. Now imagine that the cloud infrastructure supporting the online banking front-end is actually hosted on the same mainframe as the transactions are being crunched. Event management and workflow integrations become easier to link, resource sharing is greatly simplified, performance rocks. Makes sense doesn't it?

So that's what I am going to be working on with the fine folks at CA.

For my ECM stalwarts just understand that one day all your ECMaaS systems will be running on mainframes if the "big (unstructured) data" dream comes to fruition. Then you’ll wish you’d read all the way to the end of this entry! Thanks for humoring me over the years, it was a blast.

---

^[1] IDC, Market Analysis Perspective: Worldwide SaaS & Cloud Services, 2011: New Models for Delivering Software, #232239, Dec 2011