@thread

Someone could put up a poll of how many people would be interested to start "testing" of backwards incompatible NG MOS/AOS (and pay for it, even).

( version for me should boot on T1042RDB-PA ... )

Last edited by KimmoK on 05-May-2014 at 03:46 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

Quote:

Well I don't know of such systems but doubt greatly that 2004 hardware would have any significant overhead from using memory protection/virtual memory (those two are linked in most current architectures) but I know that for a 386 system one could expect up to 15% performance hit when enabling virtual memory. That was over two decades ago, close to three in fact.

I think this was true (and might be, although performance hit should be much lower today).

How much would you gain by fully exploiting multiple cores ?

And btw, what was the impact of the context switching that happened in classic when calling 68k/ppc code from/to ppc/68k ?

_________________
http://www.warpdesign.fr/

@KimmoK
Quote:

Different people have different view to what they consider Amiga. To me it's more about system simplicity and speed than the programming api etc.

AHhhhhh, +1.

_________________
Tinkering with computers.

@Leo

Quote:

Leo wrote: Quote: Well I don't know of such systems but doubt greatly that 2004 hardware would have any significant overhead from using memory protection/virtual memory (those two are linked in most current architectures) but I know that for a 386 system one could expect up to 15% performance hit when enabling virtual memory. That was over two decades ago, close to three in fact. I think this was true (and might be, although performance hit should be much lower today).

The overheads are much smaller today. I'll use x86 processors as examples: the 80386 didn't have caches except for page table entries (commonly called TLB or translate lookaside buffers) in order to speed up virtual memory handling. Todays processors not only have caches, they have multi-level TLB caches. The 80386 was designed for segmentation combined with virtual memory, todays processors are designed for virtual memory with most segmentation support executed either in microcode or complex micro-operations. For the 80386 virtual memory was an option given the software base (mostly MSDOS) while modern processors are designed to run with virtual memory enabled all the time - while it is possible to disable that setting is unlikely to be optimized.
Also in 64 bit mode virtual memory is _required_ by the specification.

Quote:

How much would you gain by fully exploiting multiple cores ?

Depends on what will run on those cores. Linear scaling with the amount of execution resources is (at least theoretically) possible for a small subset of tasks, even super-linear scaling could be possible but extremely unlikely.
In practice the speedup is limited by the amount of serial work needed for the task, well known as Amdahl's law in computer science.

Quote:

And btw, what was the impact of the context switching that happened in classic when calling 68k/ppc code from/to ppc/68k ?

I've not used such a system but have read that it was very expensive.

I think browsers would benefit first by multiple cores since they run many parallel tasks for downloading and decoding images and stuff in homepages. They all have to be synchronized properly and should need no modification for multiple cores. Webkit itself obviously already runs on SMP platforms.

On Classic PPC a context switch implies flushing both CPUs' caches. A very expensive operation which makes it meaningless to offload any small jobs to the PPC. Only large jobs, like decoding a complete jpeg or a substantial chunk of mp3 data makes sense.

_________________
- Don't get fooled by my avatar, I'm not like that (anymore, mostly... maybe only sometimes)
> Amiga Classic and OS4 developer for OnyxSoft.

@Deniil715

As you are aware AmigaOS4.1 Classic does not suffer from that context switch problem as the 68k CPU is only used for initial booting of the AmigaOS4 kernel. After that, the integrated 68k emulator kicks in and the real 68k cpu is turned off completely.

Just wanted to remind people of that

_________________
AmigaOS 4.x Beta Tester - Classic Amiga enthusiast - http://www.hd-zone.com is my Amiga Blog, check it out!

@HammerD

You must mean 68k cpu is used to load kernel and then machine restarts in ppc mode?

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@HammerD

Yes of course when running OS4. I was certain he meant the AMP model used by PowerUP/WarpOS, since we were discussing multi-processing.

_________________
- Don't get fooled by my avatar, I'm not like that (anymore, mostly... maybe only sometimes)
> Amiga Classic and OS4 developer for OnyxSoft.

@Deniil715

Yeah, Odyssey would be even better if it supported multiple cores :) But also some heavier programs like Hollywood would greatly benefit from multiple cores. (to clarify: yes, Hollywood isn't that heavy by itself, but the more complex things you do with it ...)

_________________
Proud user of AmigaOS 4.1 on an AmigaONE 500. This is the first Amiga I've ever had so I don't know all the ins and outs of AmigaOS yet, so I'm sorry if I'm asking noob questions and stuff.

@Vistaus

I wonder if it is possible to create threaded apps with hollywood for non-amiga platforms today?


@AMP

As some work tasks require combination of SMP and AMP to scale well for more than 8 cores, there might be some interesting playground/work for PowerUp and WarpUP gurus on future Amigalike systems....

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Quote:

As some work tasks require combination of SMP and AMP to scale well for more than 8 cores,

Well in her example she adds a Mutex, to synchronies the results, so the 24 threads has to wait and threads become serialized when delivering the results.

But when she runs use 8 cores per WxWorks, there are less Mutex serialization.

If the calculations did take more time, the over head of the final Mutex want be not so noticeable.

If she run the programs 3 times using 8 cores instead of running WxWorks 3 times, she might get just as good result, because then there want be 24 threads fighting over the results, only 8.

Quote:

there might be some interesting playground/work for PowerUp and WarpUP gurus on future Amigalike systems....

what a odd thing to say, PowerUP was made by Phase5, and the company is out of business. WarpUP was created by Hagge&Partner, and company has stopped supporting Amiga, they sold most of there stuff to an other company. But anyway PowerUP and WarpUP was solutions created by companies that did not have access to AmigaOS source code at the time.

AmigaOS was unaware of PowerUP/WarpUP this is way the context switch was needed, not rally any type AMP or SMP as I understand it.

Even with source code of AmigaOS, there is going to issues implementing SMP/AMP, anyway some form of MultiCore support should be possible, just because its hard to implement true SMP or AMP, in any case it should be possible to put the 2en core to use. And this is what Hyperion is working on whit X-Kernel.

http://blog.hyperion-entertainment.biz/?tag=amiga

Last edited by NutsAboutAmiga on 09-May-2014 at 12:17 PM.
Last edited by NutsAboutAmiga on 09-May-2014 at 06:08 AM.
Last edited by NutsAboutAmiga on 09-May-2014 at 06:07 AM.
Last edited by NutsAboutAmiga on 09-May-2014 at 06:04 AM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@NutsAboutAmiga

>If she run the programs 3 times using 8 cores instead of running WxWorks 3 times, she might get just as good result, because then there want be 24 threads fighting over the results, only 8.

Yes. That was only one example, not any real application.

>>there might be some interesting playground/work for PowerUp and WarpUP gurus on future Amigalike systems....
>what a odd thing to say, PowerUP was made by Phase5, and the company is out of business.

But most of those developers are alive. And PowerUP developers even active in our community.

>But anyway PowerUP and WarpUP was solutions created by companies that did not have access to AmigaOS source code at the time.

Yes. That's why I think the AMP implementation could now be taken to next level.
(things like seamless clustering over multiple heterogenous systems would be cool, so far clustering needs good amount of manual work to set up for the user, per application even)

>AmigaOS was unaware of PowerUP/WarpUP this is way the context switch was needed, not rally any type AMP or SMP as I understand it.

Not going to debate about words, but for me it was a AMP system.
And to my understanding, some amount of context switches are unavoidable.
With homogenous systems and access to OS sources the context switch penalties can be minimized.

>Even whit source code of AmigaOS, there is going to issues implementing SMP/AMP,

Sure.
But IMO, it's easy when one drops the (full) legacy compatibility.

>anyway some form of MultiCore support should be possible, just because its hard to implement true SMP or AMP, in any case it should be possible to put the 2en core to use.

I do not think there is anything in the way of SMP implementation, when the smart thing is done and some...all legacy compatibilitity is removed or built into sandbox.
(and the sandbox must be separate project, to not to slow down OS evolution to true NG)

(I think that is the thing MOS team is doing.)

It would be ok for me if I must reboot to legacy PPC OS when needed, as long as I would be able to taste the true NG system. Much later we will have all we need on the NG (on the 64bit+MP+SMP environment).

To my understanding MOS team will co-operate with AROS people to have legacy support (janus UAE etc) on their NG MOS.

(perhaps, I read too much from MOS people comments, but...)

>And this is what Hyperion is working on whit X-Kernel.

I think you read a lot more out of Hyerion info than what I can...
But perhaps that's just because you know more about SMP details than I.


Gallium3D Update
- Must be re-entrant, thread safe and multicore capable
X-Kernel Update
-Removed reliance on data structures (e.g. ExecBase task lists and ThisTask pointer)
-Moving scheduler to run on auxiliary cores
-All cores schedule tasks independently
-Load balancing between cores


To me it does not say it is not SMP etc.
And it does not say if it runs existing AOS4.1 SW at all.

(I'm ready to pay for NG SMP+MP+64bit AmigalikeOS, even if it only looks and feels like an Amiga etc... (with very few apps) untill SW gets ported and sanboxes built.)
I'm not ready to wait for 10years before I have those modern basic goodies.

Last edited by KimmoK on 09-May-2014 at 07:58 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

I have not heard of "cooperation" (how I would understand that) between Aros and MorphOS devs. In the past MorphOS devs used parts (components/sources) from AROS and gave back changes (how the rules require). So in a way it could be called a kind of cooperation. I assume same will happen in future if the MorphOS devs think it is good for them. I have read a comment from one of the MorphOS devs that they would make a clean break (drop compatibility) when they would change ISA (what was not decided yet and perhaps still not is) and 68k would be emulated. That could mean a similar or identical solution to Aros.

You say you would pay for a "NG SMP+MP+64bit AmigalikeOS" but that would mean that everything people now like at AmigaOS would have to be dropped. This would certainly be similar to AROS, not possible to simply combine 68k and native libraries to add missing features, 68k software would run in a box. Another problem, what happens with PPC software where no sourcecode is available. Or you think of staying on the PPC platform? But even if, Hyperion tries to make money with the (in my view) slowly shrinking AmigaOS community. How many people would be willing to pay for such a modernized (but incompatible) OS that can only run on a very limited number of systems? I do not think that Hyperion would be interested except someone else pay them lots of money for it and takes all business risks.

@OlafS25

> that would mean that everything people now like at AmigaOS would have to be dropped.

Simplicity, familiar folder structure, the looks, the speed... I think all most important things (for me) can be kept.

>Or you think of staying on the PPC platform?

PPC is ok (my brain is big endian).
But x64 could be an OK alternative. (ARM not so...)

>Hyperion tries to make money with the (in my view) slowly shrinking AmigaOS community. How many people would be willing to pay for such a modernized (but incompatible) OS that can only run on a very limited number of systems?

Why would it need to run on any more limited number of systems than it currently does?
Because drivers need to be rewritten? (they anyway need to be one day)

Because not all HW is 64bit? (recompile, if not possible, then forget 32bit, future is 64bit ... 4.1.6 is ok for my needs on SAM440 anyway ...)

>I do not think that Hyperion would be interested except someone else pay them lots of money for it

It would be very interesting to see a cost estimation of SMP+MP+64bit branch.

>and takes all business risks.

Perhaps it also could seen as "get some easy money" from grazy people like me.
(it could be sold as BETATESTER_AOS5_ALPHA without any warranty to any direction, and I would still love to try it...)

IMO:
-MOS has the legacy compatible amiga "DONE", they should start the NG now.
-AOS4 should not play catchup, but jump to NG now (or perhaps see if the implementation can be recompiled also as 4.1.x compatible).
-neither AROS should play catchup. 68k variant could be kept if seen sensible (important for RetroComputer/Retrogaming -products derived from AROS), but NG should be targetting to modern features on 64bit, highly multicore HW.

Last edited by KimmoK on 09-May-2014 at 10:20 AM.
Last edited by KimmoK on 09-May-2014 at 10:18 AM.
Last edited by KimmoK on 09-May-2014 at 10:16 AM.
Last edited by KimmoK on 09-May-2014 at 10:12 AM.
Last edited by KimmoK on 09-May-2014 at 10:11 AM.
Last edited by KimmoK on 09-May-2014 at 10:11 AM.
Last edited by KimmoK on 09-May-2014 at 10:04 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

"grazy people like me"

LOL

If it could be organized similar to AROS with its different branches it might be possible. But I think a ISA change would be much more complicated then adding another PPC OS. Parts of AmigaOS were licensed and I would guess that they only licensed it for PPC (then a ISA change would require rewrite and/or additional license fees). Then certainly parts are optimized for PPC (written in assembler) which have to be rewritten then too. So most propable solution seems to me would be add another PPC based OS (no ISA change). But we will see what will happen when the magical "4.2." appears

@KimmoK

I do not know if the interest in a NG OS that has different concepts (68k running in a box and no simply combining of native and 68k libraries possible) is really that big. AROS already has 64bit (I think the RAM limit is now at 128 GByte), the devs are now experimenting with SMP, it has Gallium support already for ages but how many of the AmigaOS or MorphOS user have really changed to it? What I mostly hear (besides untrue comments like it is unstable and similar) that it not offers the "Amiga feeling". Would that not be true for such a modernized but not compatible NG OS too?

Simple example is AREXX. On MorphOS and AmigaOS you can add support by simply copying the original library (as I understand and similar to what I do with Aros 68k). The library is closed and code is lost. The author seem to have vanished (I tried to find him a couple of times). Perhaps even dead who knows. I assume that even Hyperion has no sources (I think he was not paid for his work by Commodore). On AROS there is a reimplementation (but of course not 100%) as alternative (that I f.e. have included in my 68k distribution). At the moment you have (on AROS) Arexx ports in emulation and on X86/ARM... but you cannot use Arexx ports from 68k applications in Icaros desktop (here X86) and the other way round. That is possible on PPC because both (PPC and 68k) can be combined. So a ISA change would bring benefits but it also means a different feeling (compared to PPC). So i am not sure how many AmigaOS user would accept this and pay for it. MorphOS is different because it is not as commercial as AmigaOS and AROS of course is opensource.

Last edited by OlafS25 on 09-May-2014 at 10:42 AM.
Last edited by OlafS25 on 09-May-2014 at 10:41 AM.

If they wanna go for the easy way of adopting SMP, they should just drop legacy support (up to some degree, of course) and make it an optional download from their website. Once you install it, SMP gets disabled entirely or disabled for the specific old apps support only.

_________________
Proud user of AmigaOS 4.1 on an AmigaONE 500. This is the first Amiga I've ever had so I don't know all the ins and outs of AmigaOS yet, so I'm sorry if I'm asking noob questions and stuff.

@OlafS25

> AROS already has 64bit (I think the RAM limit is now at 128 GByte),

And I should get myself to try it...
(too long time I last time got AROS experience)

>the devs are now experimenting with SMP, it has Gallium support already for ages but how many of the AmigaOS or MorphOS user have really changed to it?

I think some people that have tried have had trouble getting their HW supported (I only now have a compatible x64 HW to spare).
Some people were amazed about the amount of preinstalled stuff, but got disappointed when 10% of them worked straight out of the box (my experience).
Etc.

I think the biggest problem of AROS when I tried it was the lack of maturity. And there was something that did not feel right... perhaps because the old AROS was more like plain WB3.1 rather than WB3.9+, and I'm too used to 3.9+. ?

>What I mostly hear (besides untrue comments like it is unstable and similar) that it not offers the "Amiga feeling". Would that not be true for such a modernized but not compatible NG OS too?

Perhaps "Amiga feeling" sould be made high priority rather than things like legacy API compatibility.

I must report back about my "feeling" when I next time try AROS.

(btw. on every NG HW, if I see standard MS-DOS like textual printouts when the system boots, 50% of the Amiga feeling is already lost, Amiga had graphical early boot display already in 1990, that kind of should be redone+modernized. yet another reason to have custom HW with our own wickedly fast BIOS etc.)

BTW. let's not turn this thread to yet another CPU ISA change thread (amigalike OS can already be run on every sane ISA anyway). We have more important matters to handle for now.

Last edited by KimmoK on 09-May-2014 at 10:49 AM.
Last edited by KimmoK on 09-May-2014 at 10:48 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

64bit means 64bit nightly. Icaros is still based on X86 and that means limit is 4 GByte. As far as I know it is still ahead there because f.e. MorphOS only supports 1 GByte or 1.5 GByte (do not know the situation on AmigaOS). Theoretical most of what you mention could also be done on existing PPC hardware base (X1000/X5000). The question is would the majority of X1000/X5000 users accept it and pay for it and for which price. Something I cannot judge. But we will perhaps know more when 4.2. is available. I am personally more interested in 68k (and all related developments) so I do not care much about it . I personally do not think that any of our platforms will ever becoming mainstream again (how many seem to dream of). That would even be the case if we would have 64bit/SMP/MP and lots of drivers. Even if you only have what is already available on Windows/Mac/Linux and only a fraction of the components/software that is available on these platforms. Amiga is for fun, for serious work you have plenty of options. And we need "geek" solutions that might attract people outside the community. FPGA based devices comes there to my mind. Competing with other platforms for "serious" applications has no chance.

Even Linux has not a big standing in the "desktop market" that is dominated by Windows.

Last edited by OlafS25 on 09-May-2014 at 11:06 AM.

@Thread

Read the thread subject and think again before posting

_________________
Philippe 'Elwood' Ferrucci
Sam460 1.10 Ghz
AmigaOS 4 betatester
Amiga Translator Organisation