Open system versus mainframe stability.

[Binamra]

New question as suggested here: How to know the IP address of the mainframes?

Just a different question, if this is how redundancy works, then all open systems to which the most of the companies plan to move can implement 'redundancy' to give the high availability like mainframes? But most of the times such migration does not work as good as mainframes, why?

[Robert Sample]

Your question about why open systems have problems with high availability is one with multiple answers. The first answer is that IBM mainframes are designed and engineered for high availability -- components all the way up to the processor level are designed to be able to fail over to an alternate component; this is not always the case with open systems. The second answer is that high availability is not cheap -- and open systems are not built to make lots of money. The third answer goes to architecture -- system z machines are quite capable of running with 100% CPU utilization for days at a time (at one of my previous employers, month-end processing ran the CPU to 100% for about 3 to 5 days each month) whereas Intel servers (for example) tend to start having issues when they get much over 20% CPU utilization and when they get over 80% they are close to dead as far as doing work. The fourth answer is the software -- system z software works in a multiprocessor multiLPAR environment without any issues; sometimes getting open systems software to just acknowledge another system can take some tweaking.

The bottom line is that open systems can be set up for high availability but it requires special work to get them to do so; system z machines are ready for high availability almost out of the box (as I said in the post, the SPOF have to be analyzed and resolved, sometimes before buying the machine).

IBM announced the System/360 in April 1964. In the 54 years since, each new generation of hardware has been carefully designed to maximize reliability and minimize system downtime. Perhaps when open systems have been around for more than 50 years, they will approach the reliability of the z13 and z14 machines.

[vasanthz]

Few months ago, when we closed the door on the Mainframe box, the power cable of one CPU fan would get dislodged and turn off. When one fan stopped, the other immediately picked up speed and was able to cope up with the unexpected failure. End users never noticed a thing. Mainframe kept running until the power cable was repalced by IBM a few days later.

It was impressive to see redundancy in action on Mainframe.
I suspect they have that on Intel systems. I don't know.

[Binamra]

Hi Robert,

Thanks a lot. Is not it that IBM also has processors from Intel too?

[Robert Sample]

IBM may have processors from Intel (I don't know), but I do know that the mainframes IBM builds use custom, proprietary chipsets. From the z14 Technical Guide Redbook (SG24-8451) page 4:

IBM continues its technology leadership with the z14 server. The z14 server is built by using the IBM modular multi-drawer design that supports 1 - 4 processor drawers per CPC. Each processor drawer contains five or six Central Processor (CP) single-chip modules (SCMs) and one Storage Controller (SC) SCM. Both SCMs are redesigned by using 14 nm FINFET SOI technology. {3} Each CP SCM has 10 processor units (PUs, or cores). In addition to SCMs, CPC drawers host memory DIMMs, connectors for I/O, oscillator interface, Flexible Service Processors (FSPs), and cooling manifolds.

and from page 13:

CPC drawer
Up to four CPC drawers are installed in frame A of z14 server. Each CPC drawer houses the SCMs, memory, and I/O interconnects.
Single Chip Module technology
z14 servers are built on the superscalar microprocessor architecture of its predecessor and provide various enhancements over the z13. Each CPC drawer has two logical CP clusters and one SC SCM. Two CPC drawer sizes are available in z14, depending on the number of CP SCMs. The z14 model M05 has six CP SCMs that include 49 active cores. All other z14 models have five CP SCMs that include 41 active cores. The CP SCM has 10 cores by design, with 7, 8, 9, or 10 active cores, which can be characterized as CPs, IFLs, ICFs, zIIPs, SAPs, or IFPs.

and on page 14 the processor speed is stated at 5.2 GHz.

So if you are thinking a mainframe is just a souped-up Intel machine, you need to completely change your thinking. IBM uses its own chips that bear little to no relationship to Intel chips.