Hacker News Reader

Comment by Athas | Hacker News Reader

Can recommend this link; it's a one-click victory in buzzword bingo:

> As businesses adapt to capitalize on digital, trust will be the currency that drives this new economy.

> With z14 you can apply machine learning to your most valuable data to create deeper insights.

> In the digital economy data is the differentiator. IBM z14 can rapidly derive actionable insights and enable progressively smarter decisions for better customer experiences and new revenue streams.

> IBM z14 can help you accelerate development and delivery of secure, scalable services with new economic models.

With such key features as:

"100% encryption is 100% mainframe"

"Trust – the foundation of digital relationships"

"Facilitate new secure services with Java"

I actually like IBM mainframe tech, but this page is hilarious.

View on HN

The one important part in all the buzzwords:

> New software pricing tied to business value

So, it's like a tax on your business, if you succeed, so do they!

All joking aside, has anyone used mainframes for a greenfield project in the last 20 years? I'm genuinely curious. I've been under the impression the only reason they're still around is because of the massive 20 Million LOC cobol apps hanging around.

I've often thought I'd much rather have a cheap, featureful "walled garden" PAAS for side and small projects, as opposed to wrangling vms and containers all the time. Mainframe could fit this use case.

They make some capacity and performance claims when it comes to modern software:

>> it can run the world's largest MongoDB instance with 2.5x faster NodeJS performance compared to paltry x86-based platforms. It supports 2,000,000 Docker containers and 1,000 concurrent NoSQL databases.

Don't you use mainframes to have one very powerful machine? Like the central system of a large bank that needs to ensure transactions for millions of accounts book correctly.

If you run VMs anyway, aren't x86 server cheaper? (honest question, don't know much about that)

Mainframes used to be a lot about concurrent I/O for transaction processing - e.g. booking/check-in/handling of airline passengers or yes, batch transactions in the millions (e.g. end of day calculation within banks or financial information providers).

TCO might be lower with x86 these days, it all depends on your workload.

One possible advantage is faster communication between nodes. If you want to run 2 millions docker containers, probably you'll run few hundred different servers.

If you choose 86 servers, then you need to setup all VMs cluster management, backup, recovery, migration..

I wonder who would actually use a proven inferior, troubled document store (MongoDB) on a high-end, state-of-the-art, really expensive mainframe.

>> New software pricing tied to business value

> So, it's like a tax on your business, if you succeed, so do they!

You may jest, but the flip side of this seemingly buzzword laden sentence in the feature set as a potential buyer is that our company may not have to pay if IBM fails to demonstrate incremental business value at the end of an engagement (aka conditional fees[1]).

Buzzwords can work both ways, thankfully.

[1] https://en.wikipedia.org/wiki/Contingent_fee

Mainframes are all about VMs and containers. They just use different terminology and tech.

I've heard of a few new deployments (more accurately expansions) where it's cheaper to buy more mainframe than Oracle licensing.

> Mainframes are all about VMs and containers. They just use different terminology and tech.

Exactly. The tech is called LPAR: https://en.wikipedia.org/wiki/Logical_partition

English Wikipedia link: https://en.wikipedia.org/wiki/Logical_partition

Didn't realize Linux supported LPAR?

LPARs are implemented on hardware/microcode level (some older implementations even involve actual physical switches in the CPU and IO buses for partitioning the system), so OS does not have to even know about it.

The main feature of virtualization on zArchitecture are not LPARs, but the fact that that it meets the Popek and Goldberg virtualization requirements perfectly and thus there are no artificial limits on how deep you can nest the hypervisors. In fact even the process in z/OS can almost behave as if it was running directly on hardware.

You can do virtualization also using z/VM operating system, each user has its own VM.

It still stores all the users password in plain text inside a DIRECTORY file that resides on the maintenance userids. I've seen it 100% of the time I did support on one of those systems.

Probably only truly new deployment that comes to mind is this: https://en.wikipedia.org/wiki/Gameframe

Whether it actually made sense to implement (it very well might) that way is another question.

Not sure if you would count that as greenfield projects (I would not), but I am told they make very good database and ERP servers. Our local utility runs the DB2 backend for their SAP installation on a (small-ish) mainframe.

I remember hearing a while back (~10 years) that IBM and SAP were giving fairly generous discounts for customers to move their SAP installations to GNU/Linux running on zSeries mainframes.

None of that is very buzzwordy, but putting such mission-critical software on a mainframe can make sense at a certain scale.

> So, it's like a tax on your business, if you succeed, so do they!

So... Do you realize this is basically the core of the SAAS/IAAS/anything-AAS business model ?

It is sad to see people joking at IBM just because it's IBM.

Yes. There is plenty of new development. In many places (banks, insurance) where there are old dogs that haven't learned any me tricks.

I'd avoid mainframe development as much as possible,and there is plentiful of wrangling on mainframe too...

I have a better link. Scroll down to the end, and click on "Redbooks"; there you'll find the technical information we HN readers tend to be more interested in.

In particular this one: "IBM z14 Technical Introduction" https://www.redbooks.ibm.com/redbooks/pdfs/sg248450.pdf (I'm still reading it, but page 5 already has a short list with "the main technical enhancements in the z14 over its predecessor platforms").

> Up to 32 TB of addressable real memory per system.

Imagine how many tabs you can keep open with this ;-)

Also, each processor has 128 MB of L3 cache, compared to 38 of the priciest Xeon, and an L4 cache to complete that.

No matter how much memory you have, chrome will find a way to consume it.

20-30 I guess...

Here's a diff of the z13 and z14 Technical Introductions:

https://draftable.com/compare/quHdhGnzauCm

You forgot that they mention blockchain technology.

And how!

"Blockchain with IBM Z creates unprecedented levels of trust and transparency, allows alignment with cognitive capabilities and enables deployment in the cloud."

BINGO!

Classy!

Spoke my mind.

> one-click victory in buzzword bingo:

https://vimeo.com/12112636

machine learning.. ok what ML platform would one run on it? Doesn't it make more sense to use a cloud ML provider? scale infinitely?

Or does it only apply to companies that doesn't want its data to get out to the cloud?

Yes, this is what genuinely puzzles me the most. I get that mainframes have their place for highly stable IO-heavy processing, but I've never heard anyone claim that they are particularly effective for compute-bound workloads. Especially when machine learning does not require much robustness - you just reboot the cluster nodes when they crash (and they will, because GPU drivers are crap).

Yes I thought it was commodity systems with dual xenon's and a MFT of memory and 24-36 TB DASD per node and for networking your choice of (infiniband / 10g/ 40g) for hadoop style HPC

you can of course add GPU's there are some workstation boards specifically designed for this

> ok what ML platform would one run on it?

Not sure, but it will definitely be called "Watson"

It more about the fact that tons of customer data and transaction history are already on the platform. Why move the data when you can do the science locally.

but you are bound to a huge capital investment for what is a finite performance potential.

Reply on news.ycombinator.com