@sundown

Quote:

What the hell are you on today? I though he did a great job of explaining the boot proccess.

What I'm going on about now, if you check my post, is the post-VGA process. Between the logo appearing and actual bootup my machine has a 40 second delay besides boot menu delays. So what is it doing for these 40 seconds? That's a long time in computer world.

@Signal

Quote:

Signal wrote: Why does X1000 take longer to boot? X1000 does not have a Boot ROM. Instead, the X1000 runs a program. The program is CFE, Common Firmware Environment. CFE digs around in the hardware finding devices and physical addresses for all the various components on the motherboard at switch on. It is more like a software BIOS than a Boot ROM. It is very much like booting Linux. First the Kernel gets loaded, then the Kernel boots from a fixed or assigned address. As the Kernel starts working it digs around in hardware setting up various features. You can see what it does in the logs. Then after that is done it boots into a shell or a display manager. Pretty much the same for CFE. And like the Linux kernel if new/fixed/updated features are required it only needs a rebuilt CFE or 'kernel' module. If the original NG Amigas had CFE instead of a boot ROM they could have been updated with a software download instead of having to burn a new ROM. Same for the classics. Also, no ROM 'work arounds' necessary. So it takes a little longer to boot, but it is worth the wait. Rather innovative on a Desktop Platform.

I don't have time to discuss each and every detail (there are quite a few misconceptions there), so please let me make only some general remarks which will also keep the focus on the topic.

In all, your post says that the only reason for the slower X1000 boot time is the CFE: that's not true, as it is involved only in part of the boot process.

The boot process is essentially divided in 3 phases:
* preliminary initialization of the hardware;
* loading and launching of the Kickstart;
* execution of Kickstart startup (initializations, running of startup-sequence, loading of Workbench, loading of startup applications).

Now, as detailed in my previous posts, my more-than-10-year old XE beats the sensibily more powerful X1000 hands down in all areas. CFE being slower than U-Boot isn't a sufficient reason*. There must be other reasons: we know that one is graphics card initialization; and what else? why is Kickstart loading slower (BTW: is it done in PIO mode also on the X1000?)? why is Kickstart startup slower?

*I have no idea if it actually is slower. And, anyway, CFE being slower than U-Boot wouldn't be exactly a big selling point to the end user

Last edited by saimo on 03-Jun-2013 at 05:48 PM.

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@AlexC

Quote:

AlexC wrote: @saimo The current version of CFE on the X1000 is slow but I'd say it's normal as it's not really optimized for speed, the main focus being on reliability. Some things can be optimized like cranking up the clock to 1800 so that the emulator and basic instructions can run faster, shorten unnecessary delays, possibly skip steps that aren't needed after a reset, but some things can't easily be improved much because of the way they work, like the SB600 which requires many steps to setup, you can't just dump configuration tables into ranges of registers, you have to toggle a bit here, wait, read that bit again to make sure it's set or wait some more, write an address there, write some byte(s) elsewhere, wait again, and sometimes even toggle the bit back, wait and read it yet again. Such convoluted steps only take milliseconds but they prevent you from "uploading" all the values at once so I can't begin to imagine what it takes to initialize modern gfx card. Maybe there are shortcuts, I wouldn't know.

I'm not saying that there are not sound technical reasons and/or obstacles hard to overcome. But, as the comparison between your X1000 (which is probably the fastest booting I've read of) and my XE clearly indicates that the slowness is not just in hardware initialization (see my previous posts).

Quote:

And then there's the question of who is going to spend his time applying all these tweaks? Working on that particular source code doesn't exactly look like fun.

Sure, I'm perfectly aware that the human resources are very limited - I've already said that in clear words. Actually, I've also added that I'm not blaming anybody. And, to say it all, I'd be exactly one of those who would hate to dig in that code.
That said - I'm repeating myself again - the boot performance to me seems to indicate that there are some issues which might have an impact also on the post-boot functioning of the system (how it can be that Kickstart loading and then the following startup takes longer on your X1000 than on my XE is pretty much incredible) and the excuses that CFE is slower and that modern graphics card take a long time to initialize do no good to anybody.

Last edited by saimo on 03-Jun-2013 at 05:08 PM.
Last edited by saimo on 03-Jun-2013 at 05:06 PM.
Last edited by saimo on 03-Jun-2013 at 05:05 PM.
Last edited by saimo on 03-Jun-2013 at 05:04 PM.
Last edited by saimo on 03-Jun-2013 at 05:04 PM.

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@saimo

Quote:

Now, as detailed in my previous posts, my more-than-10-year old XE beats the sensibily more powerful X1000 hands down in all areas. CFE being slower than U-Boot isn't a sufficient reason*. There must be other reasons: we know that one is graphics card initialization; and what else? why is Kickstart loading slower (BTW: is it done in PIO mode also on the X1000?)? why is Kickstart startup slower?

All areas? Memory transfer? Disk speed? CPU speed?

So far its superior in boot time, but so is C64 to all (esp. with those nice modules
As you know, computers grow complex over time, so unless OS is stored in ROM
like with phones and tablets, its needs time to boot.

CFE is surely slower then UBoot, and surely there is room for
optimization in what? About second minor OS version released for X1000 (OS 4.1.5 / OS 4.1.6)

Seems that boot time is no longer priority as there are major components to work to and OS grows more complex, even that should be mark of Amiga style.

Remembering how it was in OS 4.0 development version
http://www.youtube.com/watch?v=yp0ukoM_rNQ
and also used to beast 4 years ago on old A1
http://www.youtube.com/watch?v=T7ofkdlGGF4

_________________
Future Acube and MOS supporter, fi di good, nothing fi di unprofessionals. Learn it harder way!

@vox

Quote:

vox wrote: @saimo Quote: Now, as detailed in my previous posts, my more-than-10-year old XE beats the sensibily more powerful X1000 hands down in all areas. CFE being slower than U-Boot isn't a sufficient reason*. There must be other reasons: we know that one is graphics card initialization; and what else? why is Kickstart loading slower (BTW: is it done in PIO mode also on the X1000?)? why is Kickstart startup slower? All areas? Memory transfer? Disk speed? CPU speed?

The context is and has always been exclusively that of the boot process. In fact, the mentioned "previous posts" reported comparisons of the various boot phases and, also, what preceded the part that you quoted made it crystal clear.

Quote:

So far its superior in boot time, but so is C64 to all (esp. with those nice modules As you know, computers grow complex over time, so unless OS is stored in ROM like with phones and tablets, its needs time to boot.

This is irrelevant to this discussion, because both my XE and the X1000 boot the same OS.

Quote:

CFE is surely slower then UBoot, and surely there is room for optimization in what? About second minor OS version released for X1000 (OS 4.1.5 / OS 4.1.6)

I must admit I don't fully get what you mean.
Anyway, once again, I can't see why again the focus is restricted to just CFE. My XE loads and runs the Kickstart faster than the X1000 (I've grown tired myself of repeating this).

Quote:

Seems that boot time is no longer priority as there are major components to work to and OS grows more complex, even that should be mark of Amiga style.

Maybe it's not your priority, but it is to me and others (as expressed in this very thread), and in particular developers, because the lack of complete memory protection and resource tracking forces annoying reboots. Not to mention that everybody would appreciate a fast boot (especially the Amigans who, guess what, are the most interested in the system).
That resources are limited and thus make optimizing the boot process hard is no question. But saying that the boot speed is not a priority is just an excuse, which, moreover does no good as it leads to ignoring, f.ex., the meanigful fact that the Kickstart starts off more quickly on an old machine than on the latest and most powerful of all.

Last edited by saimo on 03-Jun-2013 at 05:53 PM.
Last edited by saimo on 03-Jun-2013 at 05:46 PM.
Last edited by saimo on 03-Jun-2013 at 05:44 PM.
Last edited by saimo on 03-Jun-2013 at 05:37 PM.

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@Hypex

All of us feel the pain, I think 50 sec is too long & beta testers have bent a few ears as well. My mistake was starting this thread here instead of on the Hyperion forum. The data here will be lost, it needs to be in a place for the developers to see & get a good idea whats going on & fix it.

Personally, I don't sit around cold booting my system all day. My SAM440 boots faster, but feels locked up many times compared to the x1000. So, if you want fast cold booting, get a SAM440, if you want a fast system, use the x1000. I beta test & I don't need to cold boot often enough to let it bother me.

One thing we all know, no matter how fast they get it to boot, someone will complain, so why bother?

_________________
Hate tends to make you look stupid...

@sundown

Quote:

One thing we all know, no matter how fast they get it to boot, someone will complain, so why bother?

I can think of at least 4 good reasons:
* the faster, the less people will complain;
* the faster, the less stressful the work of developers and betatesters;
* a fast boot is an appealing feature to the vast majority of users (I'd be surprised if there were anybody who wouldn't appreciate or at least deem better a quick startup);
* understanding why the booting takes long might lead to discoveries that are beneficial to the OS as a whole - no guarantee of course, but, hey, I can't repeat enough that if the Kickstart starts up less quickly on an X1000 than on an ancient machine, then it could be a sign that something in the Kickstart is wrong, which should be taken into careful consideration, given that, something wrong in kernel might well affect the whole system during its normal operations.

Last edited by saimo on 03-Jun-2013 at 08:11 PM.
Last edited by saimo on 03-Jun-2013 at 08:10 PM.

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@saimo

Quote:

This is irrelevant to this discussion, because both my XE and the X1000 boot the same OS.

True. As we see:

- CFE could boot faster. Maybe some newer version could do initialization faster.
- Graphic card could initialize faster: Seems this depends from card to card
- Kickstart could load faster: Either X1000 Kickstart is more complex, or disk transfer could improve. Development of better driver for onboard X1000 SATA surely could help.

With all 3 areas covered, surely X1000 could boot closer to older Terron boards, if not faster.
Sadly only developers and beta testers can do this kind of improvements.

Quote:

I must admit I don't fully get what you mean.

That OS4 for X1000 is still young. As we all know, X1000 was ment to be sold with OS 4.2
and not OS 4.1 for X1000 is stop-gap to have machine sold earlier and drivers developed,
as well as usually OS 4.2 is a bit late. But hopefully reachable

_________________
Future Acube and MOS supporter, fi di good, nothing fi di unprofessionals. Learn it harder way!

@vox

Quote:

True. As we see: - CFE could boot faster. Maybe some newer version could do initialization faster. - Graphic card could initialize faster: Seems this depends from card to card - Kickstart could load faster: Either X1000 Kickstart is more complex, or disk transfer could improve. Development of better driver for onboard X1000 SATA surely could help. With all 3 areas covered, surely X1000 could boot closer to older Terron boards, if not faster.

Is it even allowed to think that an X1000 *could* boot faster than an XE?

As for CFE and graphics card initialization: already discussed.

As for the Kickstart being more complex: apart from some specific drivers, it must be roughly equivalent to the the XE's (I guess that the Kickstart on the X1000 delivers the same functionalities of the Kickstarts running on other machines). Kickstart complexity does not seem a likely factor to me.
Also the size shouldn't be that much different (I've already asked: how much is the overall size of an X1000 Kickstart? - without considering the unused modules, of course), although it would have an impact if modules were loaded in PIO mode.

As for disk transfer: my XE loads the Kickstart in PIO mode from a PATA HD connected to the a1ide.device... it doesn't take much to beat that
Regarding post-boot transfer, I'd find it really hard to believe that my XE outperforms the X1000 when it comes to drive DMA transfers, even if the X1000 drivers aren't super-optimized. And if that is indeed the case, then there we have exactly a key area of improvement for the whole OS.

Quote:

Sadly only developers and beta testers can do this kind of improvements.

Quote:

That OS4 for X1000 is still young. As we all know, X1000 was ment to be sold with OS 4.2 and not OS 4.1 for X1000 is stop-gap to have machine sold earlier and drivers developed, as well as usually OS 4.2 is a bit late. But hopefully reachable

While it's true that the limited number of developers and the young age of the system have an impact on this all, still they can't be taken as a justification to what seems a poor performance (note: I'm not saying the developers do so). I'm thinking mostly of the fact that the Kickstart, after having been loaded, should naturally start up faster on the X1000, thanks to its faster CPU (despite only 1 core is used), faster RAM, faster drive access (I assume it is). I really believe that that's well worth investigating.

Last edited by saimo on 03-Jun-2013 at 09:20 PM.

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@saimo

Kickstart is 17.4MB & doing a lot more then on the A1's. Why do you keep trying to compare the 2 systems? Another reason, the x1000 starts up at 500MHz until the boot option comes up, means CFE is running slow at startup when the gfx card init happens. Heard CFE is a bitch to work on, so low priority at the moment.

_________________
Hate tends to make you look stupid...

@saimo

Quote:

While it's true that the limited number of developers and the young age of the system have an impact on this all, still they can't be taken as a justification to what seems a poor performance (note: I'm not saying the developers do so). I'm thinking mostly of the fact that the Kickstart, after having been loaded, should naturally start up faster on the X1000, thanks to its faster CPU (despite only 1 core is used), faster RAM, faster drive access (I assume it is). I really believe that that's well worth investigating.

With all things you have mentioned, yes its well detected problem, but surely from here it can only go faster. CFE can be written to improved version, Kickstart and OS themselves could be optimized more for the system, and yes OS 4.2 will surely utilize Nemo better.

In other areas as far as I remember just for precaution Kickstart now loads a lot of drivers. A system tool that could optimize it for users config as simply as commenting out needed parts would be nice. As far as I do remember, this use to help quite a lot on all other systems.

Truly, hardware side sais it should be faster, so it seems its just to CFE and OS optimization.

_________________
Future Acube and MOS supporter, fi di good, nothing fi di unprofessionals. Learn it harder way!

@sundown

Quote:

I think 50 sec is too long & beta testers have bent a few ears as well.

I think I would just make ears go red. But I retested and found my VGA delay is worse than thougt but my overall boot time is better. It gets to the Workbench in just over a minute without menu delays.

My VGA delay is almost 20 seconds. Setting up hardware after that takes 15. Almost 17 to load and render Kickstart. And about 16 to load AmigaOS.

I notice those post VGA 15 seconds are spent setting up USB and then reading drives according to my serial log.

Quote:

My mistake was starting this thread here instead of on the Hyperion forum.

Maybe, but on there it has a chance to be locked.

You could still do that.

Quote:

So, if you want fast cold booting, get a SAM440,

I have no need. I have an A!-XE which is faster than a 440! Which I use daily.

Quote:

One thing we all know, no matter how fast they get it to boot, someone will complain, so why bother?

That someone could be me. We should bother because we should be concerned about it. The X1000 is a big financial investment and it asks a lot. Sure a lot of work and other finances have been put into it. But the end user wants to feel like they've spent good money, not wasted it, and be rewarded. How would one feel if they could have spent quarter the money on a computer that boots and runs four times as fast? We are a tight community so getting things optimised and stable should be a priority. I don't see a small community as an excuse to be sloppy and careless if that was the case. That would push people away.

Last edited by Hypex on 05-Jun-2013 at 03:51 PM.

@sundown

Very rushed answer, sorry... total lack of time...

Quote:

Kickstart is 17.4MB & doing a lot more then on the A1's.

Good to see more information coming

17.4 MB is about twice the size of my Kickstart, so please let me ask:
* does that include *all* the modules, or just those actually needed?
* could you post a listing? I'm curious to see the file sizes.

Such double size would explain the slower Kickstart loading if PIO mode is used also on the X1000.

Quote:

& doing a lot more then on the A1's.

What exactly? This is interesting.

Quote:

Why do you keep trying to compare the 2 systems?

I'm surprised you ask, since you know the whole story
Anyway, here are the reasons:
* in the other thread, various X1000 boot times were mentiond and I was negatively impressed by the fact that they were 2-3 times longer than my XE's - pretty natural that I compare the two systems;
* both system run the same OS, so the comparison is useful to locate the shortcomings and bottlenecks on the X1000 - this is good for the OS.

Quote:

Another reason, the x1000 starts up at 500MHz until the boot option comes up, means CFE is running slow at startup when the gfx card init happens.

This is a trick you have mentioned yourself:

Other things I've done to speed up booting,
1. Set the system to boot at 1.8GHz, default is 500MHz.
In CFE, type show pmu, look for the set speed, if 500MHz, type setenv -p STARTUP "set pmu -astate=a4;menu". Ttype menu & then F to return to CFE, show pmu should now show 1,8GHz as default, this trims about 10 sec off boot time.

If, as expectable, the frequency is one of the very first things performed, the 500 MHz mode has no impact in this discussion.

But I don't get why every time the discussion is brought back to CFE and video card initialization (and we can add Kickstart loading to the mix as well): those factors have been identified already and we know that improvents are possible there.
Instead, let's focus on the fact that the Kickstart executes the startup (so, after having been loaded) more slowly on the X1000 than on the XE: that's the most important aspect (as for why, I won't repeat myself again: please refer to my previous posts). Your answer above starts to shed some light... please add more!

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@saimo

Quote:

* could you post a listing? I'm curious to see the file sizes.

Sorry, beta system & not allowed, but only a handfull are different from the A1's.

Quote:

Such double size would explain the slower Kickstart loading if PIO mode is used also on the X1000.

udma6 is default, just the opposite of uboot, you need to set the variable to slow it down.

Quote:

If, as expectable, the frequency is one of the very first things performed, the 500 MHz mode has no impact in this discussion.

Not true, the system is running at 500MHz for the the first 27 sec of my 52 sec boot time. I'd guess it would boot about 10 sec faster if CFE started at 1.8GHz. The 500MHz start frequency was considered "safe" to start with.

As far as the gfx card init, you'd have to ask Hans about his driver, he's well aware of the issues. He's done a great job, considering he had very few docs, if ant, for the cards.

Don't put to much into this, first step is to show boot info to help show slowdowns, plus this is a minor issue with more inportant 4.2 projects.

_________________
Hate tends to make you look stupid...

@sundown

Quote:

sundown wrote: System boot times have come up in an X1000 thread, I feel this issue needs its own thread, so here it is, post your boot times & any info on what you have done to reduce the time. X1000: 4850 gfx card, sata boot HD 17 sec to a-eon splash screen 27 sec to boot option menu 52 sec to WB

As the title is "Amiga hardware boot times" and I think my towered A4000 PPC qualifies as an Amiga, here is my boot time:

47 seconds.

EDIT:
This is from power on until being online.
I have MiamiDX (full version) in WB-Startup and once it is online a script retrieves the actual time from a time server on the web.
Once the proper system time is set I consider the booting process being finished.

To reduce boot time I disabled the onboard IDE with the no4000ide - HW-dongle.

System:
Towered A4000 PPC (50 mHz/200mHz - 128 mB) with Mediator, Voodoo4 GraKa, 10/100 mBit NIC, Terratec 512i digital soundcard. Deneb USB 2.0 highspeed, 256 mB-ZorRAM, ...

Last edited by Dandy on 05-Jun-2013 at 12:58 PM.
Last edited by Dandy on 05-Jun-2013 at 11:34 AM.

_________________
Ciao

Dandy
__________________________________________
If someone enjoys marching to military music, then I already despise him.
He got his brain accidently - the bone marrow in his back would have been sufficient for him!
(Albert Einstein)

@wawa

Quote:

wawa wrote: @delshay ... for a big box amiga user the best solution is to get deneb and make your own custom kick including noide-module, given you have your hd on another faster controller (like the deneb itself for example, you can boot from usb thumb). this removes the necessity of reboot and the system starts within few seconds with all bells and whistles. ...

Tell me more, mate!
Currently my HD is on the UW-SCSI of the CSPPC.

_________________
Ciao

Dandy
__________________________________________
If someone enjoys marching to military music, then I already despise him.
He got his brain accidently - the bone marrow in his back would have been sufficient for him!
(Albert Einstein)

@sundown

Quote:

Sorry, beta system & not allowed, but only a handfull are different from the A1's.

How about someone post a list from a public release?

Last edited by Hypex on 10-Jun-2013 at 02:01 PM.
Last edited by Hypex on 07-Jun-2013 at 03:35 PM.

@sundown

I appreciate the new information you have shared, but I have no time to comment now (hectic days).
Without breaking the NDA, could you please answer also my other questions/remarks?

_________________
RETREAM - retro dreams for Amiga, Commodore 64 and PC

@Dandy

Quote:

As the title is "Amiga hardware boot times" and I think my towered A4000 PPC qualifies as an Amiga, here is my boot time:

Comparisons with other systems always helps.

_________________
Hate tends to make you look stupid...