@broadblues

No. Very much not.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@umisef

Your avatar almost looks like a NUMA.

Quote:

In other words, NUMA means that some memory is "closer" to one processor, and other memory is "closer" to another processor

So this is a bit different to $100 being closer than $200. From that description NUMA looks like it is dependant on hardware rather than being fully in software. So not only cores but other processors neded as well.

I can only imagine that making it "NUMA aware" would be done in order to ease the transition into multicore support somehow. Right now they are thinking of how to work around Forbid()/Permit() issues so access to cores can be arbitrated. But if those functions are a problem it's their fault for sticking them back into the new OS! And removing/corrupting other functions. Compared to running 68K interrupts inside an actual PPC ionterrupt. Who does that!?

Another ideas is that they tdon't want to be stuck behind again. In case the NUMA oppertunity shows up.

Also wonder if an Amiga with a PPC card meets NUMA regulation? Though AFAIK both CPUs cannot be ru nin tandem so no multitasking across CPUs.

But, what about a 2x dual core G5 Apple, four cores total across two CPUs?

@agami

Quote:

They did indicate that they would be running two (2) instances of Exec, and the two Execs would play nicely with each other by keeping to their memory address space.

I asked about this some years ago. And from I was told Exec doesn't really run on the CPU. It runs inside an interrupt. So all this sharing woulkd have to be interrupt based. As that is what controls the whole operation.

@Hypex

Quote:

looks like it is dependant on hardware rather than being fully in software

NUMA is a property of hardware, yes. On non-NUMA hardware (like all OS4 hardware present and announced), NUMA support in the OS will do exactly nothing. On NUMA hardware, memory accesses are always non-uniform. If the OS is aware of it, it may be able to make some choices that improve the fraction of fast accesses, and thus improve overall performance. If the OS is not aware of it, things still work perfectly fine, just not necessarily as fast as they could.


Quote:

But, what about a 2x dual core G5 Apple, four cores total across two CPUs?

Alas, no. That was my guess, too, but then I checked out the hardware block diagram. The memory isn't attached to the processors --- which is how you typically end up with NUMA; The processors contain memory controllers, and obviously memory attached to a processor's controller(s) is closer to that processor than memory attached to another processor's controller(s).

In the quad-G5 the two dual-core PPC970 are hooked up to a system controller, which then controls the memory. Thus, the distance between each processor and each memory is exactly the same --- one CPUsystem controller link, plus one system controller memory link.

@Kronos

Quote:

Kronos wrote: @cdimauro Since you don't have a law-degree your not even allowed to enter that discussion

You made my day!

@agami

Quote:

agami wrote: @cdimauro Quote: NUMA and SMP definitions cannot be changed as someone wishes We're not changing the definition of NUMA. We are speculating or making educated guesses as to how a NUMA-aware kernel would benefit AmigaOS 4.x running on a single socket multi-core CPU with interleaved/non-interleaved RAM.

Since such machine is clearly NOT a NUMA one, it doesn't make sense to implement a NUMA-aware kernel, which can even slow down the task scheduling.

@umisef

Quote:

umisef wrote: So --- anyone want to speculate why the Friedens are making ExecSG NUMA-aware?

...or it's merely the usual smoke/mirrors and part of the constantly moving mirage called "OS 4.2" that we have seen for the last decade?

Quote:

cdimauro wrote: Quote: Kronos wrote: @cdimauro Since you don't have a law-degree your not even allowed to enter that discussion You made my day!

+1

I see NUMA as being the next level above SMP. As at some point there is a diminishing return per added CPU core, due to maximum bandwidth of the main memory RAM bus there, they are then trying to get the "sweet spot" of SMP performance into a cluster, and then making multiple of those clusters. So that SMP main-memory kindof starts to look like an L4 cache (if you have an L3 cache CPU core) covering that whole cluster, and another cluster has its own L4 main-memory level, etc. And then there is some capability to span from one L4 to another, which is the "remote" or "distant" memory concept, accessing the main memory of a different cluster. Then, all of these L4 segments together make up the total "main memory" before you get to further and slower hard disk swap memory. It adds a level of hierarchy between what we think of SMP cpu/memory and the hard drive swap.

t4240 as mentioned elsewhere in this thread, has 3 SMP-alike cpu core clusters, and 3 DDR3 memory controllers. Each cluster is best on it's own memory controller, but has capability to access memory on a different controller, at a lower performance level than its own. That's my understanding today. I hope to be learning more about this thing in the near future.

_________________
All glory to the Hypnotoad!

@umisef

Since AmigaOS seems to be stuck in 31-bit land for all eternity (the entire AmigaOS is apparently dependent on the MSb being zero everywhere for the system to run, according to an answer elsewhere), they could kind of fake NUMA in an SMP setup by mapping the low 2GB to one core and the high 2GB to another and thuse use all 4GB in SMP "NUMA" mode

This setup (having two low 2GB virtual spaces in on 4GB phy space) would then favor one core to use one halv of the RAM and the other core the other half. Any cross access would mean copying or remapping or "bank-switching" or something.

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

Quote:

Since AmigaOS seems to be stuck in 31-bit land for all eternity

Hopefully they will expand their computing horizon sometime. And get rid of 30 year old archictecture and move on so that today's (and future's) hardware can used correctly without hacks even MSDOS devs would be ashamed of...

Last edited by Leo on 15-Sep-2015 at 09:31 PM.

_________________
http://www.warpdesign.fr/

@Deniil715

Don't think that count as NUMA either, and any way the ExtMem objects map into the top 2-4Gb part of the memeory space (so must only be used with caution with any OS related routines).

_________________
BroadBlues On Blues BroadBlues On Amiga Walker Broad