Feeds

back to article Mainframe emulator goes commercial

Roger Bowler - the creator of the open source Hercules mainframe emulator - has put together a company called TurboHercules to try to commercialize the decade-old program that he created as a "programmer's plaything." Rather than go straight at the IBM mainframe base, which many a company has tried to do and ended up in court, …

COMMENTS

This topic is closed for new posts.
Gold badge
Thumb Up

A massive achievement

Emulating a whole mainframe architecture (and its predecessors) to the extent it can hosted on a totally different architecture hosted on *multiple* other OS's.

By 2 main developers.

There is an epic saga to be written here.

Incidently I think the space Shuttle;s General Purpose Computers are essentially S/370 architecture with a few additional instructions (IIRC most of the additions do not work). AS they use no known IBM OS re-hosting the software may be a bit tricky.

0
0

hercules is great fun...

... to play with. Spent many a happy hour playing about with MVS and pascal/fortran 2. Always was a bit disappointed never to find z/OS on piratebay.

There is aslo a lovely alpha/DEC emulator out there for running openVMS as well if you're fed up playing with linux.

B.

0
0
Go

Slow?

I've used Hercules a little as a way to get Fedora working on S/390, and it seems to work well. 'Course I don't have a £ half million mainframe to compare it against.

One problem is that it's pretty slow - reputedly about 20 times slower than the real hardware. So it doesn't sound like much of a disaster recovery solution, unless they're going to address the problem of emulation speed.

0
0
Anonymous Coward

IBM must hate this

IBM seem to have expended a lot of energy in trying to squash this over the years. IBM internal employees have been banned from using it, instead having to use the fiddly, expensive, dongle-encumbered Flex-ES product and being threatened with the sack if Hercules was used. Didn't stop a lot of people at IBM playing with Hercules on the quiet and booting z/VM and z/OS on it however.

A shame that z/OS isn't licensed as free-to-use for non-commercial purposes. Might end up getting people interested in playing with the mainframe platform. As it is, I gave up because it was so closed and unwelcoming.

0
0
Grenade

This goes to --By John Smith 19 --

If you do a video search you will find that the space shuttle is being supprted by Amiga500 computers as the hardware and OS is opensource and is VERY highly configureable thus keepsing costs down greatly.

So I stick my toungue out at you and your fancy that Nasa supports IBM when you are flat out wrong.

0
0
Bronze badge

OS/360 is not copyrighted, though

Due to a bit of an oversight at IBM, OS/360 wasn't copyrighted. But you can only get the binaries for that, and not the source code. Oh, well.

0
0
Thumb Up

@AGirlFromVenus

More than just the alpha is available - the completely free 'simh' emulator will emulate a huge variety of hardware, including Vax, and easily runs openVMS as well.

Also, Bull open-sourced Multics a year or two ago, and I believe there are projects ongoing to produce an emulator capable of doing it justice.

0
0
Silver badge
Unhappy

Is that the sound....

....of IBM's wallet creaking open? I predict another PSI-style legal attack followed by a swift acquisition, killing and burial. IBM does not let anyone mess with their mainframe bizz.

0
1

Various points..

@ Matt Bryant: the nice thing is that Herc is open-source; it can't be killed or buried.

@ AC: The other indication of how IBM feel about Hercules is in how they handled the 'Redbook Saga'. IBM produced a Redbook on how to run Linux on S/390. This writer made use of Hercules in figuring out how to do this, and the book made quite a few references to Hercules. There was an entire chapter dedicated to running Linux on Hercules, for those who couldn't afford a real mainframe. A few months later, a revised version of the Redbook was issued. Silently; contrary to normal practice IBM did not increment the version number of the book. The revision consisted entirely of the systematic excision of any reference to Hercules, including the whole chapter. Go figure. Or see: http://www2.marist.edu/htbin/wlvtype?LINUX-VM.25658

0
0
Grenade

Is it really viable?

You may be able to use the licence for a DR site *in the event the mainframe fails*. So any attempt to plan, develop, and test the other site by starting up <licence program code> on some cheapo box sounds like it would break the licence terms. Cos your mainframe hasn't failed. That loophole is as frail as it sounds.

IBM can poison the well if necessary through hardware, software, or licence terms changes. Probably some DMCA bollocks in there as well to 'leverage' against mis-using code.

The tactic of appearing very open and co-operative, non-threatening to IBM is a nice approach for PR, since when IBM acts it will look unfair and bullying.

If its worth having DR, if you have to have a mainframe for the RAS features or the parallel IO options, you don't necessarily want to bring it all up on a completely different infrastructure with different procedures and probably without any vendor support from IBM. When your primary data centre is a smoking heap of rubble you might want someone to answer the phone. This does not seem like freetard territory.

If the EU attempts to act, IBM could pull products or support. Is the EU going to do that to all the banks etc that use it? Economic self-harming ahoy.

AC cos...you can probably guess.

0
0

1 IBM Mips equals 4MHz x86

According to a Linux expert, 1 Mips == 4 MHz x86. Thus, a Mainframe CPU is not really that fast. A Nehalem is probably faster.

http://www.mail-archive.com/linux-390@vm.marist.edu/msg18587.html

It is suggested that you can replace Mainframes nowadays with a Linux/Unix server. You can emulate a Mainframe on your laptop with Herkules.

http://blogs.sun.com/jsavit/entry/once_again_mainframe_linux_vs

Mainframes were really fast in the days of intel 286. But today, a Nehalem is pretty fast. Next year a Nehalem with 8-12 cores at high frequency will be released. It will probably crush a Mainframe CPU. A Mainframe with 8 sockets cost millions dollars. And will be outperformed by a decent x86 Nehalem server running Linux or Unix at a fraction of the price.

IBM dont like to release general Mainframe benches, why is that if the Mainframes are so good? Probably the benches show lousy performance. IBM braggs when there is something to brag about. I would like to see IBM's cheapest Mainframe against a fast clocked Nehalem EX with Herkules. I bet the Nehalem will win.

0
1
Bronze badge

What is needed

IBM was once forced to unbundle - to sell software separately from hardware. I don't suppose that it really dominates any market to the extent that it could be forced, by antitrust law, to treat x86 and Itanium simulations the same as competitive mainframe hardware.

I think it's a pity that 360-architecture chips aren't commodity parts with the same price/performance as x86 chips - and that they, rather than the baroque x86 chips, aren't dominating the desktop.

0
0

@Kebabbert

The Mainframe forté is IO rather than CPU.

0
0
Bronze badge

@John Smith 19

Whilst producing a mainframe emulator in software is an achievement, it's not actually that difficult compared to producing something like a data base or a fully-fledged operating system (although make it run fast is tricky). The original 360 & 370 architectures are very simple with an extremely regular instructions set. The original POP (principles of operation) was really quite a small document.

I rather suspect a competent C programmer could write a basic 360/370 processor emulator in a couple of weeks (although it might be rather slow). It used to be very common for emulators to be written before any new processor was released to give the operating system developers something to work on (that was in the days when computer companies designed new CPUs).

Of course the architecture has move on a lot since the days of 24 bit address spaces, and various extensions (and not a few kludges) have been added, but writing processor emulators is a relatively easy thing to do (it's emulating devices which get difficult, although, fortunately, mainframes are not rich in the number of different devices that you can plug in, so it's not a difficult problem). If an emulator is to exploit multiple CPUs fully then that also make life tricky, but it can be done.

There are hardware emulators out there for pretty well every obsolete machine you care to name (I've got one for an ancient machine called a Nascom II that had a Z80 processor which I built from a kit).

As always, the trouble is with access to software. Legacy apps are the IBM's largest sales tool (and legacy support people). Yes, the hardware is nice and reliable, as is Z/OS, but they are also riddled with unpleasant software architectural issues which are left over from the past. Nevertheless, a nice little earner for IBM, especially given many of their customers in this are are conservative and highly risk averse.

0
0

hercules is not so platform-limited..

The article states that "32-bit x86, 64-bit x64, and Itanium processors", when in fact it is not limited to those architectures at all. I am personally running it on PowerPC and UltraSPARC architectures, and will be building it on ARM9 later this week. It is reasonably portable code.

Just FYI.

0
0

Re: 1 IBM Mips equals 4MHz x86

Actually (referencing Kebabbert's comment), it's not likely that a Nehalem can come anywhere near the performance of a modern mainframe. People have been predicting the death of mainframes for thirty years, and they're still what keep the world turning.

Perhaps some information on the systems-level architecture would be illuminating.

An IBM z10 can be configured with sixteen quad-core processors clocked at 4.4GHz, with 3MB of L2 cache per core and 24MB of four-core-shared L3 cache. These processors have four 13GBps memory ports, two 48GBps SMP hub ports, and two 17GBps I/O ports. There are often as many ancillary processors (for handling I/O, accelerating Java, etc) processors as "regular" processors in such a system, depending on the configuration.

Another important aspect of mainframe design is reliability. There's a great deal of internal redundancy in these machines...you can walk up and put a bullet into one of these racks and, chances are, you'll never know it unless you go look at the OS error log.

This is not PC hardware; there are reasons they cost millions. Be glad your bank transactions won't be handled by Nehalem machines.

0
0
MDR
WTF?

Mainframe MIPS

Um... 1 MIPS = i million instructions per second.

0
0

clockspeed?

c'mon people, as if clockspeed is the only reason for companies to buy system z hardware...do you really think any company wouldn't protest? Unparallelled IO bandwith for instance: hundreds of applications running simultanuously, all doing IO... up to 1,5 TB of memory... give me the name of the itanium box that delivers that, please? Just read about system z a bit, IBMs websites deliver enough information.

0
0
Headmaster

Re: 1 IBM Mips equals 4MHz x86

Dude, where do you even come up with this stuff?

Before you continue with that line of argument, please read the following links. I think you vastly underestimate what mainframes can do. If you had read some specifications, you might know that the contest you suggest implies pitting a cheap Nehalem server against a machine with several four-core z6 chips, each running at 3.5-4.4GHz and connected to a system bus running at a few gigahertz and more RAM than your computer likely has hard drive space.

1. http://www.serverwatch.com/hreviews/article.php/3737101/Server-Snapshots-IBM-z10-EC.htm

2. ftp://ftp.software.ibm.com/common/ssi/pm/sp/n/zsd03005usen/ZSD03005USEN.PDF

3. http://speleotrove.com/decimal/IBM-z6-mainframe-microprocessor-Webb.pdf

More important, however, than the muscle to replace an entire data center's worth of commodity servers from Sun or Dell (which, granted, many mainframes can provide) is the reliability of a mainframe and its ability to operate continually under extremely heavy load for years at a time without ever screwing up, even in the event of component failure. These cheap Nehalem servers running Linux--do they have hot-swappable RAM and processors? If so, how many, and how fast is the I/O underneath them? Moreover, how well do they respond to new tasks if they are already under 80% CPU load? Part of what makes mainframes nice is not just their computing power, but how effectively they use it, and how long they can keep up with that kind of strain without ever stopping.

Is it really so foolish to spend one million dollars on a computer which is capable of maintaining the entire IT capabilities of a company whose business is worth a hundred times that, in order to ensure that it will never bog down, crash, or need to be taken down for maintenance, even if it's at 80-95% load for a decade? No cheap Nehalem server is going to manage that. An entire data center worth of them (and expensive ten-gigabit ethernet switches) might, though it would cost nearly ten times the electricity, have far higher rates of component failure, have vastly slower I/O between the individual machines than the mainframe would manage between its internal VMs, and generally be a lot more expensive to keep running than the mainframe. For that matter, is it foolish to buy two or three $1M mainframes, so that the first can be clustered with another offsite?

Please read up on this stuff before making unverified claims about it.

0
0
Grenade

Re: Kebabbert

"According to a Linux expert, 1 Mips == 4 MHz x86. Thus, a Mainframe CPU is not really that fast. A Nehalem is probably faster.

http://www.mail-archive.com/linux-390@vm.marist.edu/msg18587.html"

Haha That data is 6 years old, and 2 generations out of date. How many Niagara processors does it take to beat a single Nehalem?

"It is suggested that you can replace Mainframes nowadays with a Linux/Unix server. You can emulate a Mainframe on your laptop with Herkules.

http://blogs.sun.com/jsavit/entry/once_again_mainframe_linux_vs

Mainframes were really fast in the days of intel 286. But today, a Nehalem is pretty fast. Next year a Nehalem with 8-12 cores at high frequency will be released. It will probably crush a Mainframe CPU. A Mainframe with 8 sockets cost millions dollars. And will be outperformed by a decent x86 Nehalem server running Linux or Unix at a fraction of the price."

The blog you linked, and your comments are just hand waving. Coincidentally, Jeff Savit spent an awful lot of time recently working on solaris for z :)

"IBM dont like to release general Mainframe benches, why is that if the Mainframes are so good? Probably the benches show lousy performance. IBM braggs when there is something to brag about. I would like to see IBM's cheapest Mainframe against a fast clocked Nehalem EX with Herkules. I bet the Nehalem will win."

Haha Sugar daddy Oracle had to persuade Sun to do clustered TPC-C

All vendors play games with benchmarks. Why do you cheer lead for Sun?

0
0

Reread my posts???

I have posted links to an Linux expert showing that 1 MIPS equals 4 MHz x86 (in the days of Pentium4). Read that link again. Nehalems are much more powerful than a P4, clock for clock. I've also argued that the Mainframe CPUs are not that fast in comparison with a fast Nehalem. Say that a normal Mainframe CPU does 10.000MIPS (costing 100K USD or even more?).

Consider a Pentium 4 at 3.2GHz with 8 cores. It has 8 cores x 3200MHz = 25,600 MHz. Tthat corresponds to 6,400 MIPS, Now, how much more powerful is a Nehalem than Pentium 4, clock to clock? Twice? Thrice? If you look at the benchmarks on TomsHardware of a 3GHz Pentium 4 and pit it against a 3GHz Nehalem, the Pentium 4 maybe gets around 30 points in a typical benchmark, and the Nehalem gets maybe 250points. Almost 10 times more. Let us be generous and say that the Nehalem is only 3 times as fast as a P4, clock to clock. Then 25,600MHz would correspond to 19,200 MIPS. That is more than 10,000MIPS to me. If a Nehalem is 5 times faster than a Pentium 4, then it would correspond to 32,000MIPS, which is faster than any Mainframe CPU. If it is 10 times faster than a P4, then it corresponds to 64,000MIPS. Add Herkules ontop and let us see the cheapest Mainframe costing a million USD? vs a fast Nehalem EX machine with software emulation, costing 5K?

.

Captain Thyrathron

"Dude, where do you even come up with this stuff?"

Nowhere have I talked about the I/O of a Nehalem. I have only talked about raw power of the CPUs. Havent I? Have I lied to you? Have I not provided links and hard facts? You could email that Linux expert that "debunks the Mainframe myth" in that link Ive posted if you wish to know more on his work. If you claim that he lies, then I would like to know where he lies. If you are correct, that he is a liar, then I will not write such things again.

But then you have to prove that he is wrong. And your IBM propaganda doesnt count. IBM states that the Power6+ CPU has 250 GB/sec band width, but IBM adds all bandwidth in the CPU. That is clearly false, if there is a bottleneck with 10GB/sec then there will be no higher bandwidth than 10GB/sec. IBM people are not technically knowledgeable, or they are twisting their propaganda. Pick your choice.

.

Why do I always cheer for SUN? Well, they are open source. And they have the best tech. ZFS rules. And DTrace. And Niagara. And Zones. etc etc The list could go and on. I only support the best tech. I am geek. If IBM gets the best tech, I switch.

0
1

Time then

Time then to dust out my copy of XT/370 then .......

0
0
This topic is closed for new posts.