Slashdot Mirror
IBM Launches Linux-Only Mainframes
An anonymous reader writes: IBM is introducing two mainframe servers that only run on Linux. It's part of a new initiative from the Linux Foundation called the Open Mainframe Project. "The idea is that those companies participating in this project can work together, and begin building a set of open source tools and technologies for Linux mainframes, while helping one another overcome common development issues in the same manner as all open source projects." IBM's hardware release is accompanied by 250,000 lines of code that they're open sourcing as well. "Ultimately the mainframe mainstays are hoping to attract a new generation of developers to their platform. To help coax new users, IBM will be offering free access to the LinuxOne cloud, a mainframe simulation tool it developed for creating, testing and piloting Linux mainframe applications." Canonical is working with IBM to bring Ubuntu to mainframes.
Mainframes are nice in that you get hw with 100% uptime. Not 99.99%, but 100%. Electronics getting old? Need replacements? Offline a couple of CPUs, then pull the cards while the machine is running. Insert new ones and bring them up. Repeat, until you've swapped all the CPUs - and the mainframe was running all the time! (Obviously not at 100% capacity, but transactions were processed continuously).
Memory modules are hot-replaceable in the same manner. So is network, disks & power supplies. All is redundant, all is replaceable without shutting down. You can do such stunts "to some extent" with PC hardware - i.e. you can get a pc-type server board with redundant power. And linux has hot-adding of CPUs already. But mainframes has 50 years of experience with this sort of always-up requirement - so it just works, without snags.
Modern Z hardware has nothing to do with POWER. Mainframes do not push computation into channel controllers, whatever gave you that bizarre idea? Treat your database like a device driver? What is that supposed to mean? Linux runs native on zSeries, so virtualization is not necessary (and has not been for more than a decade).
You seem to know absolutely nothing about mainframes, why are you posting?
Nokia's DX200 series of PSTN switches had fully redundant motherboards, you could literally physically cut the PCI bus and the thing would just keep on rolling, without dropping any calls..
And that's where Dell will come in and put thousands of cores in a 42U rack for you...
We're getting to the point where all that matters is how much performance can you get from an assemblage of nodes, and how much does it cost to buy and support it?
If IBM can provide a lower TCO than Dell with different technology and the "containers" are compatible, many customers will be interested.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
Mainframes are not simply overpriced PCs. They're put together internally in quite a different way.
The original system busses were in the backplane, not in on a "motherboard". That was true even on my very first (S-100) PC, long before IBM got into the personal computer market. The backplane was almost nothing but wiring, with no caps to blow. You'd basically have to set it on fire to render it useless.
Z series and power definitely do not share an instruction set, and they have really substantial differences, but that isn't keeping the engineering teams all that separated, if indeed they are at all.
Quoting Timothy Prickett Morgan from http://www.itjungle.com/tfh/tf... , "And as has been the case in the past, the Power and z processors are designed by a single processing team and are borrowing technologies from each other. This does not, however, mean that IBM is creating a converged processor that can support either Power or z instruction sets." My hazy memory makes me think they're sharing FPU blocks, possibly one of the bus interfaces, and it seems like one of the cache blocks (L3?). Z has plenty of custom hardware - I think it's fair to say it's predominantly custom - the branch predictor would have to be pretty different, and of course power doesn't have a BCD arithmetic unit.
Point being, if you're going down the Z Series road to run a Unix-like OS, why not just (conceptually) stop early, end up with something like Power, and call it good? Anyway, I'll argue that they're spiritually and economically related, and there's more than a passing family resemblance. Kind of like power and modern ("advanced server") iSeries, though that's getting more into Deliverance territory.
Meanwhile, channel controllers aren't as dumb as they look. A little wikipedia action here (I know, citing wikipedia, but it's monday and I'm still tired): https://en.wikipedia.org/wiki/... . Turns out the little dickens can do a decent amount of work on its own. I think the wikipedia entry is showing its age... seems like IBM's done a lot more work since this.
I remember when SASI came out. I looked at the spec and thought "Hey, this is a lot like a channel controller." Then I read some more and decided "No, a channel controller is much smarter. But this isn't bad." SASI became SCSI and everything else flowed downhill from that. At a very real level, Linux is forcing a million dollar fibre channel array to look more or less like an ST506 connected a board from 1984. Wild.