Oracle Switching Solaris To A Continuous Delivery Model

[ldo17]

And you seem to make distinctions which are logically defying themselves.

The difference between a “supercomputer” and a simple “cluster” is all down to the speed of the interconnect. “Embarrassingly parallel” problems are those that scale up linearly (or close to linearly) with the number of CPUs available--for example, rendering frames of a CG animation. More complex problems--like weather forecasting--have heavy needs for communication between the CPUs, so they don’t run well on simple clusters. That’s where supercomputers come in. And that’s why they cost more.

And all the most powerful supercomputers in the world run Linux.

[aht0]

Network-based interconnect yields delays measurable up to milliseconds range. Regardless of the speed and nature of the connection . Processors sharing memory bus have delays measurable in low nanoseconds. Look up the differences between nano, micro, milli.

Difference. You can use one for number crunching where delays do no matter. Engineering- and scientific problems. Simulate nuclear explosion in 3D for example. 'Real-time' is not important.

Other is used for speed critical stuff, transaction processing for example, where your networked supercomputer's responses would remain way too sluggish. How much 'power' you supercluster has does not matter in the least..
Clear?

[linuxgeex]

Hm, you ever heard of IBM z Systems?

LOL, yes. And obviously you have something to share. So share. Links to platform specifics, SAP marketing materials, and/or current benchmarks will make your words meaningful.

To be clear here, I didn't say Power was currently the most expensive system. Only the most performant.

Here's the current top SAP certified systems:
Tier 2 (single system) Fujitsu / Sparc 844420 SAPS
Tier3 (cluster) IBM / Power 1471680 SAPS

Note that these are 2013 systems. There's 2017 systems in the current SAP certified benchmark tree but none beat these 2013 systems yet.

The fastest uncertified z13 machine I could find with google-fu approaches 50,000 SAPS, that was a 2015 model, there's no faster current model for sale.

[drSeehas]

... Links to platform specifics, SAP marketing materials, and/or current benchmarks will make your words meaningful.

Your IBM sales representative will be happy to bombard you with marketing material.

To be clear here, I didn't say Power was currently the most expensive system. Only the most performant.

You wrote "most powerful server IBM is selling". Maybe we have a different understanding of "powerful".
There must be a reason (e.g. Five nines) why customers pay more money for the more expensive and more "powerful" z Systems. Don't you think so?

[linuxgeex]

Your IBM sales representative will be happy to bombard you with marketing material.


You wrote "most powerful server IBM is selling". Maybe we have a different understanding of "powerful".
There must be a reason (e.g. Five nines) why customers pay more money for the more expensive and more "powerful" z Systems. Don't you think so?

lol, yes, see your own words above "Your IBM sales rep..." but that doesn't make z Systems more performant or more reliable, it just means that they're doing a good job of selling support, and that support will help guarantee that the person buying the hardware doesn't do stupid things with it. Commodity x86 hardware can achieve five 9's if the stack is architected and serviced appropriately. There's no secret sauce. Reliability comes with a price tag because it requires forethought, redundancy, and perseverance by a well trained, experienced Devops team. IBM has that, and brand awareness going for them. On that, you and I obviously agree. :-)

[aht0]

lol, yes, see your own words above "Your IBM sales rep..." but that doesn't make z Systems more performant or more reliable, it just means that they're doing a good job of selling support, and that support will help guarantee that the person buying the hardware doesn't do stupid things with it. Commodity x86 hardware can achieve five 9's if the stack is architected and serviced appropriately. There's no secret sauce. Reliability comes with a price tag because it requires forethought, redundancy, and perseverance by a well trained, experienced Devops team. IBM has that, and brand awareness going for them. On that, you and I obviously agree. :-)

Linux does AFAIK not guarantee backward compatibility to business software written 20-30 years a go. It has in fact trouble being backward compatible to itself. And such old software is still being pretty widely used. For example by credit card companies, each time you swipe your credit card, it in all likelyhood is ending up touching some mainframe somewhere and some of the code being executed is pretty fucking arhaic. z Systems do. It's a also a kind of reliability.

Why IBM is moving to using Linux. Pragmatic reasons. Mainframe sysadmins (or 'system programmers') are increasingly retiring, there is no replacement pool, since IBM did not think on donating mainframes to the universities back at the time. New generations of sysadmins know only Linux and x86, nothing about mainframes. So, it's a sort of compromise. Best of only "bad" choices.

[linuxgeex]

Linux does AFAIK not guarantee backward compatibility to business software written 20-30 years a go. It has in fact trouble being backward compatible to itself. And such old software is still being pretty widely used. For example by credit card companies, each time you swipe your credit card, it in all likelyhood is ending up touching some mainframe somewhere and some of the code being executed is pretty fucking arhaic. z Systems do. It's a also a kind of reliability.

Why IBM is moving to using Linux. Pragmatic reasons. Mainframe sysadmins (or 'system programmers') are increasingly retiring, there is no replacement pool, since IBM did not think on donating mainframes to the universities back at the time. New generations of sysadmins know only Linux and x86, nothing about mainframes. So, it's a sort of compromise. Best of only "bad" choices.

Yes, some telco's are still running Multics lol

[ldo17]

Linux does AFAIK not guarantee backward compatibility to business software written 20-30 years a go.

It certainly guarantees backward compatibility with old code written for Linux. Linus has famously chewed out people who wanted to break userland compatibility, on more than one occasion.

[aht0]

It certainly guarantees backward compatibility with old code written for Linux. Linus has famously chewed out people who wanted to break userland compatibility, on more than one occasion.

Have you actually tested your grand theory?

I've ran into issues starting from compiler errors (which is pretty fucking logical and expectable), ending with a needed functionalities being already removed from the kernel.

[ldo17]

... needed functionalities being already removed from the kernel.

Such as?