A short glimpse on OS 2.8

[Mike Kostousov]

Does anybody know about "/proc/sys/pm/suspend"? What is it? Trying to read from this file puts iLiad to sleep. Don't ask me how to wake up May be there is a problem?

[scotty1024]

Originally Posted by Malder1

I also owner of iLiad (maybe you remember I wrote utility that renders text to images for PDF to reach better sharpness of text for reading)

Now about to be obsolete with new pdf display technology I helped develop for the iLiad.

[scotty1024]

Can we take a deep breath and then let it out slowly here?

Power will get better. I've read the PXA-255 CPU manual, I've poked and prodded. I'm not worried as I see so many things to change for the better its like being a kid in toy store with Daddy's Black Amex Card.

The code base is very fat and bloated and much can still be done at the application layer.

Today I've removed nearly all the X server code from ipdf. It now writes directly to the frame buffer and bench marks show that I'm consistently decreasing memory foot print and CPU usage as I pare down the code.

Poppler is next. As I put in the iRex required error distribution routine(s) I'll code them, as well as the zooming logic, using the PXA-255 CPU's MMX opcodes to increase their efficiency.

In addition, right now benchmarks in ipdf show it takes Poppler 1.5 seconds to 2.6 seconds to draw a single page of just text (like say my Baen RTF encoded as PDF 1635: Canon Law). That is an outrageous amount of time and thus I know there is something just waiting to be optimized in the text rendering logic. Ditto my Jack Rabbit Performance Test, a simple CCITT 2D encoded bitmap takes 30 seconds to decode on a 32 bit 400MHz PXA-255? I don't think so! A frickin' FAX machine running an 8 bit CPU at 16MHz can do it faster than that!

Things are going to get better.

Just remember, iRex didn't write Poppler. And the Poppler writers didn't write it for a hand held. Nothing personal in all this, it just needs to be re-worked for the new reality it finds itself in.

We need to encourage iRex to continue collaboration and let us knock off a few things that are higher on our priority list than they are on iRex's so iRex is free to spend more resources on their B2B priorities.

[CommanderROR]

@Scotty1024

I agree with what you said, but...
iRex said they aren't going to give us root access...and that probably means you won't get to play around with the PM features...
And, iRex apparently actually told somebody that they CAN'T support Suspend mode...which would mean hardware limitations, like the PageBar not being wired up correctly to wake the device up or something along those lines...

I'll wait for an official Statement from iRex before I really start going mad, but the way it's looking at the moment...

Optmizing code for iRex is a very cool thing, and you have my Thanks for that (and the thanks of most other Iliad users I guess... ) but if something is fundamentally wrong with the hardware design...
Well...I'll keep hoping for the moment...

[Mambo]

Scotty:

Is there an easy to use manual on how to install your PDF reader on the iLiad?
A kind of manual for dummies? I have only a Windows PC and a G4 Mac at home, I would like to see the performance improvement you are talking about, but I am not able to hack it myself (still running 2.7 though). I would be able to do it if there would be a step by step instruction which covers everything in one document. I think it would be great help to the community...

[TadW]

Originally Posted by CommanderROR

I put this up at the iRex forums...maybe we'll finally get a straight answer on this matter...so far all answers have been a bit "unclear", so the fear and the hope both lived on...

Their official answer:

The iLiad hardware was not designed to be capable of doing a suspend-to-flash or suspend-to-ram. This implies that the CPU and the DRAMs have to be powered all the time if the user does not want to boot the iLiad every time.

Current situation:
At this moment (software v2.7.1) the processor and the DRAMs are always running at their full speeds. Also some of the circuits -like the audio sub-circuit- are not yet shut down when they are unneeded. We are working continuously to improve the power management system to achieve the maximum battery life possible.

Progress on this power management system is slow since it must be tested very thoroughly. A failure to test all cases in the power management system could result in very unstable iLiad units and that should be the last thing we want. This awareness has lead us to implement the power management system one subsystem at the time.

Battery life:
iRex tests battery life using an automated setup in which a fully charged iLiad is booted and used to read a book at 1 page per minute speed. The result of this test is reported as battery life to the users.

Conclusion:
So to add things up the iLiad will feature an advanced power management system, but its hardware was not designed to be able to suspend (or hibernate) to flash or ram. To achieve maximum battery life future software versions will throttle down the CPU to 33MHz and put the DRAMs into low power self refresh mode, when the user is reading. The iLiad will then react almost instantly when the page flipper is used.

[doctorow]

Uhm, uhm... I am very curious how they want to underclock an X-Scale 400MHz to 33MHz while keeping things stable.

[tribble]

the xscale processor is designed to change cpu speed on the fly. going fromm 33 to 400 mhz should be almost instantly.

[doctorow]

tribble, that's not my point. As far as I know the PDA255 400MHz was not meant to run at 33MHz while keeping a stable environment.

[scotty1024]

I think we're having some language issues here.

A true clock stopped "suspend" is never going to happen with DRAM.

DRAM consists of tiny little leaky buckets of electrons. If you don't go along the rows of buckets and inspect them and re-fill them they will leak down to no electrons and thus destroy their contents.

In normal high power operation the DRAMs are refreshed by the PXA-255 CPU. But in "deep sleep mode" the CPU can turn that responsibility over to the DRAMs and live out of its on board cache memory at a highly reduced clock rate of its own.

If iRex wanted to that could be using DRAM's with tighter testing specifications. The DRAM makers inspect their chips and those with the lowest leakage rate in their buckets get sold at a premium for use in devices that need the lowest of deep sleep states. DRAM with tightly picked specs can cut their onboard self-refresh rate way way way down. The normal refresh rates are selected to let the maximum number of devices operate stably. I've personally worked with DRAMs that were so low leakage they could go a minute at room temperature with no refresh and still have their contents remain intact.

The PXA-255 CPU itself is rated to drop its clock down to 33MHz and still be stable.

A "hibernate" to FLASH would indeed require a reboot. That doesn't make it impossible unless the device can't turn itself off. But with the 2.7 updater iRex showed the iLiad can do that.

The technical issues are where to get enough FLASH to hibernate to, there isn't enough onboard. So MMC, CF or USB stick. Well waking up USB stick to reocver from is not something one wants to code in a boot loader. So MMC or CF. So far though the MMC performance has been pretty slow. So then CF.

But how badly does iRex want to write a feature that only works with a 3rd party optional accessory?

Plus, as with sleep mode, there isn't any code to just go grab. Some clever person would have to code it just for the iLiad.

It could be very cool though. Suspend your iLiad onto a CF card, pull it out, put in a different CF card and resume from it. Suspend it again, put the other card back in... When tabbed readers become available this could be a very nice feature to have.

[scotty1024]

Originally Posted by tribble

the xscale processor is designed to change cpu speed on the fly. going fromm 33 to 400 mhz should be almost instantly.

It is instant in human terms but the CPU has practical realities that must be observed. There is a time delay in going from 33MHz to 400MHz that must be accounted for in the management code.

The best example I can give in lay terms is that when the CPU is shifting frequencies it is sort of like a lady changing her clothes. The CPU wants a bit of privacy. It doesn't want the WLAN card to suddenly burst in upon it during the wardrobe change.

[doctorow]

Scotty, thank you for your explanation.

There is this guy who wrote XCPUScalar and who said that unclocking the PXA255 to something like 33MHz is not feasable for various stability issues. Perhaps Windows Mobile is the culprit, but that's not how I understood it.

[segatang]

About the annonations of PDF:
Does the handwrited annonations "map correctly" after the PDF zooms +/- or rotates?

About the power management:
Discussions above seems if an iLiad user want to have maximum using time now,
he should turn off his iLiad entirely when he does not read anymore
and should reboot it with long 45 sec when he wants to read the ebook even just to take a look of 1 page?!

However, before you read a paperback book, you don't spend 45 sec to reboot, right?

The iLiad is an e-"book"; thus, in my opinion, it should be turned on as soon as possible, with whatever mode, to reserve the "basic" demand and advantage of the conventional book.

I care the "waiting time" before I read an ebook much more.
The hibernate or suspend mode should resolve both the "waiting time" and power consuming problem.

I hope the Linux OS should have the potential and possibility to overcome the hardware limitation.
Therefore, the iLiad can write the SDRAM data into the internal flash memory or external flash card to make itself into hibernate mode such as Windows does into HDD
or the iLiad can perform standby/resume mode (with very few power consuming) as the same as the X-scale CPU based PPCs or Palms can.

[wagnerian]

I just don't turn my iLiad off if once turned on.
It still runs for about 10 hours. :-)

[CommanderROR]

Oh dear...I'll open a new thread for this...

Go here to continue the Power-management discussion.
For talk about the 2.8 release, please continue in this thread.

Thanks,