Re: [PLUG] Why mainframes excel: was How the IRS is able to reverse engi

Re: [PLUG] Why mainframes excel: was How the IRS is able to reverse engineer Assembly

So I have conducted a lot of research on mainframe (and specifically MF Linux) costs... there's some things people don't really think about. For a MF, you typically get everything on a single bill except administrators, power, and real estate. All hardware, cabling, CPUs, Disks, etc.. all on a single bill. I've even seen a few companies that throw the corporate jet on the MF bill/account. Anyways, IBM claims (and I am admittedly skeptical) that you can replace a data center (size not defined) with one MF... maybe two. So you have much lower power consumption, less real estate needs, no miles and miles of cabling, less administrators, etc.. IBM argues that if you combine those with the better performance of the MF that it will cost less than the non-MF solutions. They also argue, quite correctly, that no one really knows the true cost of their non-MF solutions because it's not in a single bill but hundreds of bills making it impossible to know the true cost.

A few years ago I published an article on the problems of Linux on the MF. At the time, z/VM was the only hypervisor which was very expensive. Since then IBM has released the LinuxONE, the KVM hypervisor on LinuxONE, and LinuxONE leasing plans. All of these designed to address the steep initial cost of the MF. They did a pretty good job of addressing most of the problems my research uncovered. Has it made any difference? No idea.. I'm no longer in the Linux on z space.. I've moved over to the dark side.. z/OS.

If anyone's interested, there are a couple MF Linux user groups in the Jersey City and Manhatten areas. I can certainly share the invites.

Here's the specs on IBM's latest release of the MF, the z14. If you prefer redbooks, here's the z14 redbook. Announced back in September and not yet shipping, I think. We're in the process of installing our development system.

Performance wise... there is no doubt in my mind that it is the best platform. Certainly being a 54 year old architecture (that has evolved) there is a lot of baggage and old applications. However, some are porting to Java and Metal C, others are rewriting from scratch. Some are writing brand new applications on the MF. It is far from a dead platform.

On Fri, Feb 9, 2018 at 12:10 PM, K.S. Bhaskar <bhaskar@bhaskars.com> wrote:

I absolutely agree that hot-swap CPUs, ECC RAM, a geographically distributed system image, etc. all play to the "zero down time" theme.

Apropos performance, I do have some experience within the last decade comparing the same application (same source code, compiled using whatever gcc came bundled with RHEL on one and SuSE on the other) between x86_64 (server grade hardware, not consumer grade hardware) and zSeries, and the x86_64 was faster. Of course, there is plenty of apples to oranges in the comparison because it's hard to compare an n-way Giraffe CPU machine on one architecture with an m-way Elk CPU machine on the other architecture, but the zSeries machine was more expensive.

Regards
– Bhaskar

On Fri, Feb 9, 2018 at 2:05 PM, Carlos M. Fernández <aremmes@gmail.com> wrote:
This assumes that mainframes have been stuck with ESCON and bus&tag since the 1990s, which is far from reality. Current models use PCI Express 3.0, InfiniBand, and Fiber Channel much like x86 server hardware, and then add a ton additional technology on top of that which one rarely finds on x86 servers, e.g. hot-swappable processors and memory, RAIM, redundant PCIe, globally dispersed single system image capability.

http://www.redbooks.ibm.com/redbooks/pdfs/sg248451.pdf

And they're not limited to running legacy OSes and software. IBM will happily sell you a Linux-only configuration with KVM so that you don't even need to license z/VM for virtualization.

Best regards,
//CMFM

On Fri, Feb 9, 2018 at 12:30 PM, K.S. Bhaskar <bhaskar@bhaskars.com> wrote:
Mainframes are not the most powerful computers these days. Once upon a time (through the 1980s) they had the fastest CPU, the most memory, and the highest IO throughput. Then computer systems with CPUs like the DEC Alpha started surpassing mainframes, first in computing capability and then in RAM. Storage connected by fiber channel meant that mainframes were no longer the IO throughput champions either – everyone was on a level field. That's why, circa 2000, IBM went through a major re-branding exercise, and mainframes became zSeries (z for zero down-time), the RS/6000 became the pSeries (p for power), the System 38 (if I remember my numbers correctly) became iSeries (i for integrated), and the x86 based machines became xSeries (x for x86).

Today's champions are all x86 architecture based machines. But zSeries mainframes run programs written in the 1960s, written in COBOL, operationally managed by JCL, using databases like DB2 and IMS, and transaction monitors like CICS. For enterprises, it's often cheaper and less risky to pay the extra $$$ for mainframes than to port, rewrite, or replace the applications.

Regards
– Bhaskar

On Fri, Feb 9, 2018 at 11:40 AM, Steve Litt <slitt@troubleshooters.com> wrote:
On Thu, 8 Feb 2018 20:34:32 -0800
Lowell Higley <higleylh@gmail.com> wrote:

> Whether you know it or not, you probably touch a MF dozens of time
> every day.. airlines.. banks.. financial markets.. insurance. All run
> by mainframes. You pay social security or get a social security
> check? Yup, you guessed it.. all run on mainframes. Greater than 90%
> chance that when you use your mobile phone to check your bank account
> balance - you're touching a mainframe on the backend.
>
> An insurance company that will remain nameless for the purpose of this
> message has been trying to leave the MF for almost a decade (a M$
> alum sits on their board). They've only been able to successfully
> port one application (starting from scratch) and it's always down and
> generally performs like @(#*. Without MFs your interactions with
> many companies would be much worse than they are today (definitely
> not saying it's perfect today.) Even as "expensive" and "antiquated"
> as they are (the mainframe will turn 55 in 2019), they arguably still
> out perform and - especially when you factor in all the indirect
> costs - cost less than other platforms.

I've heard this from enough mainframe-knowledgeable people that I
believe it, so the next question is, how does the mainframe bestow so
much power? It seems to me that it wouldn't be hard to put a few
terabytes of RAM on a board. With 64 bit addressing it would be no
problem to use it all. It would also be easy to put a bunch of
processors on the system. I've heard the mainframe's real advantage is
its bus, but I haven't heard the details.

So how do mainframes derive their power?

SteveT
___________________________________________________________________________
Philadelphia Linux Users Group -- http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug

___________________________________________________________________________
Philadelphia Linux Users Group -- http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug

___________________________________________________________________________
Philadelphia Linux Users Group -- http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug

___________________________________________________________________________
Philadelphia Linux Users Group -- http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion -- http://lists.phillylinux.org/mailman/listinfo/plug