I took advantage of IBM's public mainframe system running Linux and
compiled up a copy of BitKeeper on it. Didn't have to change a single
line, just worked, thanks to gcc -Wall and friends.
This isn't a BK thing, it's a Linux thing. It's amazingly cool to me
that Linux runs on stuff as small as all sorts of embedded devices,
up to tiny PC's like a netwinder, all the way up to mainframes.
I'd really like to see the IBM guys let the walls between the linux
instances down a bit. If I could mmap the other linux instances
memory, that's a kickass system.
Anyway, kudos to the people who did the Linux/390 stuff, we'll
include it in our list of supported platforms next release.
--
---
Larry McVoy lm at bitmover.com http://www.bitmover.com/lm
Larry McVoy wrote:
> I'd really like to see the IBM guys let the walls between the linux
> instances down a bit. If I could mmap the other linux instances
I guess the wall between the images is more an advantage than a disadvantage.
And for sharing information between the images you have virtual network
drivers IUCV (VM) or hypersockets(LPAR) with a theoretical bandwith up to
24Gbyte/sec for hypersockets.
That should be enough.....
> > I'd really like to see the IBM guys let the walls between the linux
> > instances down a bit. If I could mmap the other linux instances
>
> I guess the wall between the images is more an advantage than a disadvantage.
> And for sharing information between the images you have virtual network
> drivers IUCV (VM) or hypersockets(LPAR) with a theoretical bandwith up to
> 24Gbyte/sec for hypersockets.
Some experimenting had been done with GFS across S/390 instances apparently
> I took advantage of IBM's public mainframe system running Linux and
> compiled up a copy of BitKeeper on it. Didn't have to change a single
> line, just worked, thanks to gcc -Wall and friends.
BTW grab a copy of Hercules and you can test it at home. Its a very good S390
and ZSeries (S390 64bit) emulator..
Larry McVoy writes:
> I'd really like to see the IBM guys let the walls between the linux
> instances down a bit. If I could mmap the other linux instances
> memory, that's a kickass system.
The situation is analagous to that of mapping memory between tasks in
Linux. Instead of one task mapping another's address space directly,
the usual design is to have both tasks do a shared mmap() of an
underlying file. Similarly, with multiple Linux guests under VM, you
can have them both map a shared pre-created Named Saved Segment (NSS).
Such an NSS is created by a guest with appropriate privileges via the
CP command DEFSEG and then, once mapped and primed with its initial
content, saved with SAVESEG. After that, other (permitted) guests can
access and map it into their (perceived real) address space via the
DIAGNOSE X'64' API. For details, see the manuals
"CP Command and Utility Reference" (SC24-6008-02) and
"CP Programming Services" (SC24-6001-01) available in PDF form from
http://www.vm.ibm.com/pubs/pdf/vm420bas.html
There's also DIAG X'248' ("Copy-to-Primary Service") which lets you
copy memory from the address space of another guest into your own
address space (obviously, the other guest must have set access
controls to let you do it...). Nobody has yet published nice Linuxy
interfaces to use either of these APIs but it's somewhere low down my
ToDo list and I've a feeling someone else that's been looking at this
stuff will probably beat me to it anyway.
> Anyway, kudos to the people who did the Linux/390 stuff, we'll
> include it in our list of supported platforms next release.
I'll check and see if it gets added to the Linux/390 software
availability list.
--Malcolm
--
Malcolm Beattie <[email protected]>
Linux Technical Consultant
IBM EMEA Enterprise Server Group...
...from home, speaking only for myself