Solaris for IBM Mainframes? Yes, really! (OpenSolaris for Z, ZFS, s390x GCC, and more)

Подписаться 1 тыс.

50% 1

Alright, I know folks have been asking about this one for at least a year. It took a decent bit of work (most of that centered on getting the ramdisk built properly and the VSWITCHes defined properly), but, it was sure worth it. This demonstration shows off booting the system, networking, ZFS (yes, really), compiling programs, trying to get sendmail to work, and performance profiling.
The tool I used to scoop performance information out of VM is VM Perfkit, aka PERFSVM, FCONMON, the Performance Toolkit, and whatever else it's called this week.
distribution.sinenomine.net/o...
There it is. Have fun!
Thanks Sine Nomine for making this possible those decades ago.
good A/B comparison of what my Super Professional-Sounding Newscaster Voice Simulation DSP Model 69420 sounds like at the start lol

Наука

Опубликовано:

3 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 22

@unixtohack 5 дней назад

Woow, if I see the complexity and the inner power of a mainframe … deep respect for the pro’s !!! even as retired industrial programmer ...

@moshixmainframechannel 22 дня назад

Fun one !

@BrunswickHeadsRevisted 16 дней назад

This was one of the most fun projects I've worked on. Nothing like getting the hardware to do things like DAT. We did go overboard by handling 3 levels of region tables + segment tables + page tables. If only Oracle hadn't taken over Sun we may have got much further.

@BrunswickHeadsRevisted 16 дней назад

The ABI is identical to Linux on z which made porting gcc/gdb/... that much easier. In fact I also wrote a layer that would recognize a module built on Linux to run on OpenSolaris and map its stuff (e.g. the layout of stat) so you could run Linux programs. I didn't get all that finished though. Java was going to be phase 2 as it was such a big task but that never happened. The initial java stuff was just going to use gcj.

@BrunswickHeadsRevisted 16 дней назад

The disk I/O and network I/O is slow as we made it do things simplistically as we debugged things. The limitation is our implementation not the system service. Phase 2 would've involved optimizing this stuff to make it perform.

@hs-tc 16 дней назад

@@BrunswickHeadsRevisted Thanks for your awesome insight and work, but, I just gotta ask... but, what are the chances that the in-progress Phase 2 code exists somewhere? I've combed through a source dump of Phase 1, but, the Sine distribution of OSoZ seems to be a bit more feature-laden than the source dump would lead me to be able to replicate.

@BrunswickHeadsRevisted 16 дней назад

@@hs-tc We were preparing for Phase 2 so had brought in a lot of packages we wanted built but that was it. Phase 1 was to prove: Operating System would "work" and a working gcc that passed the testsuite. I think we did get a working Apache and a couple of others but Java was untouched.

@BrunswickHeadsRevisted 16 дней назад

The disk write problem was fixed but didn't seem to make it to the public distro. It was a misunderstanding of how CP does block I/O. I had assumed that if the last block I had requested being written meant that all the others were too but there's no guarantee in what order the blocks are written. I did correct this but it appears I didn't push it.

@agle_6098 19 дней назад

Ah hellyea

@AnonyDave 22 дня назад

Oh wow, sun of that era sure managed to make platform names that confused today me. Even as someone who used solaris for serving stuff in that era, I saw you go through all the supported platforms on the system (it had all the sparc ones), and saw SUNW,Netra-T4 and thought "weird, I'm sure the netra t4 servers were several years later than that. A few minutes googling and apparently they gave that platform name to the netra 20, which was an ultrasparc 3 server. Some poor developer must've had to deal with all that a few years later when the T4 based netras came out - I have a T4-1 in my rack of obscure-ish toys 😅

@hs-tc 22 дня назад

I'll admit... I'm a bit anemic when it comes to solaris stuff >_< I'm mainly an IBM-head myself, with my Sun arsenal peaking out at an Ultra 5 and SPARCstation 5. fun machines nonetheless! the Netra T4s are serious machines for sure.

@BrunswickHeadsRevisted 16 дней назад

Trying to implement QDIO from scratch as IBM doesn't publish the details in the Principles of Operation meant we used a DIAG (CP service) written for us.

@BrunswickHeadsRevisted 16 дней назад

Also, I thought I had got dtrace running.

@BrunswickHeadsRevisted 16 дней назад

For 32-bit programs the flag is -m31 (31 bit addressing)

@grappydingus 22 дня назад

Unfortunate that this requires z/VM to run. Maybe someday, we will get this converted to standalone.

@hs-tc 22 дня назад

I won't say it'd be easy, but, I will say it wouldn't be impossible. we've got the source, and, implementing a set of (possibly ported from Linux) CCW DASD and QDIO OSA drivers is all that would need to be done to run it standalone. alternatively, it might be possible to pull a MUSIC and implement the two CP DIAG instructions it needs in herc itself.

@RhetoricaRhamnusia 18 дней назад

At 24:17, it seems you have CLFAGS instead of CFLAGS. Hence the failing make. ... I'm kind of surprised it isn't all the ^M spam that's the problem.

@hs-tc 17 дней назад

heh, two folks have said that, and, I am proud to say that that has been in that tarball since 2016!

@VK2FVAX 13 дней назад

"shutdown -y -i5 -g0 now" .. init5 will also sync the disk. Cute.. but disk I/O is woeful in dumb mode. Would've been wonderful if it was developed further to talk to channel controllers correctly.

@hs-tc 13 дней назад

yeah, it's a shame. I thought about throwing in some effort to port over some of the Linux CKD DASD support, but, I don't know if the effort to do so is worthwhile in the slightest...

@VK2FVAX 13 дней назад

@@hs-tc To me.. the better way forward would be if the kernel support code and headers for s390 were available, to try and integrate them into IllumOS project as it's still living, breathing and has users. It may garner sufficient interest from hobbyists under hercules or those handful with hardware to chip in? ..just thinking and tyre kicking.

@hs-tc 11 дней назад

@@VK2FVAX that's what I'm thinking too -- Illumos isn't totally different from 2009-era OpenSolaris, and, overlaying it might be the way to do. what I really need is the sources to the GCC cross-compiler that spits out binaries of the correct architecture and ABI.