Boot Time


You can't have a hard wired switch and still get good write preformance to your SD cards. Operating systems like linux and windows don't write out to hard disks or other removable data right away they buffer it and wait till it is less busy to write out data. This is why you need to tell OS's to shut down because they need to flush their buffers to the disk. If you don't do this you will loose data. You can rig the file system to write to the disk without buffering but you loose performance. If you hard wired the switch to the battery people would loose data or complain about slow write times.
 
Kevin_H said:
You can't have a hard wired switch and still get good write preformance to your SD cards. Operating systems like linux and windows don't write out to hard disks or other removable data right away they buffer it and wait till it is less busy to write out data. This is why you need to tell OS's to shut down because they need to flush their buffers to the disk. If you don't do this you will loose data. You can rig the file system to write to the disk without buffering but you loose performance. If you hard wired the switch to the battery people would loose data or complain about slow write times.
Oh that's right... I keep forgetting this isn't just a games console anymore. Ergh... I should just forget about the Pandora and go for something else.
 
Last edited by a moderator:
jmetal88 said:
Oh that's right... I keep forgetting this isn't just a games console anymore. Ergh... I should just forget about the Pandora and go for something else.
So, you're going to give up on it just because it takes 5-10 seconds longer to power off (The eeePC takes 5-10 roughly on power down to shut off...shouldn't take any longer than that, really...) than flipping the switch.

Fair enough, means one more for someone else... ;)
 
Last edited by a moderator:
All this talk of "under a minute" boots is scaring me a bit. Wouldn't it be possible to, uh... create a very simple OS that boots very quickly? Or to have that option, anyway? I'd rather exchange fancy GUI stuff that I don't strictly need for a quick boot-up.

I'm no expert in this at all (I own neither a PSP nor any PDA), but I do notice that Windows loads a bunch of useless crap that I don't really need when it starts... or that I only need rarely, anyway. So would it be possible to just tell some programs to not start up right when I turn the system on, if I don't plan to use them often?
 
Kevin_H said:
You can't have a hard wired switch and still get good write preformance to your SD cards. Operating systems like linux and windows don't write out to hard disks or other removable data right away they buffer it and wait till it is less busy to write out data. This is why you need to tell OS's to shut down because they need to flush their buffers to the disk. If you don't do this you will loose data. You can rig the file system to write to the disk without buffering but you loose performance. If you hard wired the switch to the battery people would loose data or complain about slow write times.
Alright, this only makes partial sense. If I were using any emulator, on any of my windows pcs, for any system, then after I hit the quick save the power could go out and I'd be safe. I should know, its happened :eek:
Well the same goes for my DS or PSP, if I do a quick save and then hard boot the device, the saves are fine.

So what is it about the Pandora that makes it so different from my ds, psp, or windows box... when it comes to handling the file writes? Even if I'm misunderstanding some of what you're saying and have good faith you'll clear it up for me in a reply, I still find it exceptionally hard to believe that if I executed a quick save in an emulator running on the Pandora and summarily powered the device off completely, that my data would be lost.

It's also possible I don't know jack. ;)
 
Last edited by a moderator:
I had this problem with an iTouch recently. Finished using it and packed it away. Came back to it a week later and it will not even power up. All because I didn't "real" power off which takes a lot longer to do.

I'll certainly use the Pandora's standby features whilst I'm using it throughout the day, but when I expect not to be using it for a few hours (incl. overnight), it'll be turned off completely, regardless of power consumption in standby mode. I do the same with my PC, Laptop and every other device.

Kevin_H speaks the truth, however, the Linux implementation isn't going to be that picky. Sure it'll cache data to speed up sd access, but a few seconds of no activity and the data will be written to the card. So unless you have only just written stuff out to the sd card, the write cache should be empty, and thus shutdown should be pretty quick.

QUOTE
I still find it exceptionally hard to believe that if I executed a quick save in an emulator running on the Pandora and summarily powered the device off completely, that my data would be lost.


That's true on PC and PSP also. If you save in an emulator and then immediately power off, you will lose data. However. if you save, wait a few seconds, and then power off, you will not in most cases. The PSP kinda guarantees this as it will not let you completely power off for a few seconds. The PC can do it in less than a second, so it usually takes you longer to kill the power than it takes Windows to save your data.
 
BTW, Squidge said it will be able to hibernate (save RAM to persistend media, aka FLash or SD) after some time suspended, thus spending "zero" battery life.
 
Esn said:
All this talk of "under a minute" boots is scaring me a bit. Wouldn't it be possible to, uh... create a very simple OS that boots very quickly? Or to have that option, anyway? I'd rather exchange fancy GUI stuff that I don't strictly need for a quick boot-up.

I'm no expert in this at all (I own neither a PSP nor any PDA), but I do notice that Windows loads a bunch of useless crap that I don't really need when it starts... or that I only need rarely, anyway. So would it be possible to just tell some programs to not start up right when I turn the system on, if I don't plan to use them often?
Heh... With all the "crap" loading, Linux, that's a slightly customized Xandros or Ubuntu, boots in 30 seconds on an eeePC, which is roughly the same class of machine overall. You should be looking at under 40 second boots as what's to be expected and probably more like 20 or so with a tuned setup.

A Palm does a full-boot in 15-20 seconds with a lesser CPU and a light OS. When you "power down" a Palm, you're actually putting it into sleep mode. Ditto a PocketPC with WinCE. A full-boot of the N800 is 20-30 seconds unless you've hacked the crap out of it with a customized install on the inside flash.

A DS or SP "boots" so fast because there's next to NO OS involved with it- it's pretty bare metal there from what I understand. Part of your game loading is what real OS might be involved with the game's operation on the handheld.

I don't know what a PSP has for boot time- I'm presuming it's got a bit of a boot time involved though.

If it boots in less than 15-20 seconds there's no appreciable advanced OS in the machine- only embedded systems firmware and it's pretty limited in what it can/can't do.
 
Last edited by a moderator:
Svartalf, We've started to use an OS at work for some of our products (media players, etc) and some of them boot from scratch (zero power to functional) within 20 milliseconds :D

I must admit though, the more complex ones (Linux OS) can take 10-12 seconds to bootup.

On the other end of the scale, we have a fairly simple network switch that runs Linux. It takes 3-4 MINUTES to bootup.
 
Squidge said:
Kevin_H speaks the truth, however, the Linux implementation isn't going to be that picky. Sure it'll cache data to speed up sd access, but a few seconds of no activity and the data will be written to the card. So unless you have only just written stuff out to the sd card, the write cache should be empty, and thus shutdown should be pretty quick.

That's true on PC and PSP also. If you save in an emulator and then immediately power off, you will lose data. However. if you save, wait a few seconds, and then power off, you will not in most cases. The PSP kinda guarantees this as it will not let you completely power off for a few seconds. The PC can do it in less than a second, so it usually takes you longer to kill the power than it takes Windows to save your data.
Thanks for all the clarification! I think Kevin meant well but it was misleading to me and possibly others to insinuate dire complications when using a hard power switch, when you take in to consideration how minimally it affects other devices. (PSP, PC).

This news about a feature thats going to dump the ram to an image after a certain amount of time sounds AMAZING. Then I could actually use the sleep feature on a device! I always turn off my psp or ipod instead of sleeping them, because I never know if I'm going to come back in 2 hours or 2 weeks. I am a little curious how a feature like this could be implemented though... it certainly seems tricky to me! For one, if there is something keeping track of how long the device is in sleep and some type of scheduler for making the image, that leads me to believe there will be more being powered during sleep mode than just the necessary wattage to keep the ram from wiping... and makes me wonder how that feature would affect average sleep time battery life.
 
Last edited by a moderator:
Svartalf said:
I don't know what a PSP has for boot time- I'm presuming it's got a bit of a boot time involved though.

If it boots in less than 15-20 seconds there's no appreciable advanced OS in the machine- only embedded systems firmware and it's pretty limited in what it can/can't do.

It takes about 8-10 seconds to do a full boot up to the XMB. (just counted :p)
 
Last edited by a moderator:
Squidge said:
On the other end of the scale, we have a fairly simple network switch that runs Linux. It takes 3-4 MINUTES to bootup.
OUCH. One has to wonder what CPU they're doing that with and what stupid thing(s) they're doing on boot...

My day job client's doing some pretty DUMB things with their ATCA cage blade software- but the switch blades only take about 30-40 secs to boot, and the diskless blades take about 60 or so seconds, all with Linux running on all of them. (We won't get into the boot times on the disk bearing blades...I did say they were doing some dumb things... ;) )
 
Last edited by a moderator:
Kevin_H said:
You can't have a hard wired switch and still get good write preformance to your SD cards. Operating systems like linux and windows don't write out to hard disks or other removable data right away they buffer it and wait till it is less busy to write out data. This is why you need to tell OS's to shut down because they need to flush their buffers to the disk. If you don't do this you will loose data. You can rig the file system to write to the disk without buffering but you loose performance. If you hard wired the switch to the battery people would loose data or complain about slow write times.
Write caching is really only useful for any device which requires seeks (ie, hard drives) or highly limited resources (ie, network).
SD cards (and other flash medium) are bloody fast compared to hard drives and have no seek time: just write and go. I've got two external hard drives and both handle write caching automatically: just write and go, let the device worry about it. To that end, I've disabled write caching for all external devices, and haven't noticed any slowdowns. As a bonus, I can just copy, paste, and pull as soon as it's done. It has the "appearance" of taking longer, in that the progress bar moves slower, but that's because before it was just preparing to write, not actually doing the write, which wouldn't happen for several second more.
As far as I know, PSP does not cache it's writes either, though it's hard to tell since you need to hold the power button for several seconds, and even then it seems to do some kind of cleanup before shutting down completely.
I can think of no reason that the Pandora should be caching whatever writes it needs to make to the cards. If it does, I can only hope it's a configurable option somewhere that I can just flip off.
 
Last edited by a moderator:
Svartalf said:
jmetal88 said:
Oh that's right... I keep forgetting this isn't just a games console anymore. Ergh... I should just forget about the Pandora and go for something else.
So, you're going to give up on it just because it takes 5-10 seconds longer to power off (The eeePC takes 5-10 roughly on power down to shut off...shouldn't take any longer than that, really...) than flipping the switch.

Fair enough, means one more for someone else... ;)


Don't pay me any attention, I'm just pissed off for various reasons concerning my real life right now. It's easy for me to get upset about nothing at the moment. But seriously, I probably won't get one, at least not at launch. I don't think it'll be worth it until it has a decent library of games built up. I don't emulate much anymore and I'm not interested in the UMPC capabilities.
 
Last edited by a moderator:
jmetal88 said:
Don't pay me any attention, I'm just pissed off for various reasons concerning my real life right now. It's easy for me to get upset about nothing at the moment. But seriously, I probably won't get one, at least not at launch. I don't think it'll be worth it until it has a decent library of games built up. I don't emulate much anymore and I'm not interested in the UMPC capabilities.
Heh... I'm in the business of making the games in question, so I guess I'm in no matter what. As for the UMPC capabilities, part of that brings much more powerful gaming possibilities than a handheld alone would bring to the table. I don't blame you for wanting to wait- it's understandable.

Sphinxter said:
If you don't like linux buffer cache mount it with a nobh. You will have to pay attention to the block size though, write less by reading block, modifying then write.
The drawback to this would be read and write performance as you force the machine to read from and write to the disk directly. We'll have to see if we can "cheat" and just simply do that- if we can make it work well off of everything that way, perhaps killing the buffer cache systemwide would be doable and then it could tolerate a straight-up power down with few, if any issues- well none more than with the current crop of handhelds.
 
Last edited by a moderator:
rumsey said:
This still sounds a little presumptuous considering you don't have the numbers for the 3530, and the random 5% guess... but it sounds like you have a little better idea behind the power distribution in the hardware during sleep mode, so could you explain why the PSP eats so much battery in sleep mode? Surely its not using some ancient nasty kind of ram that sucks down gobs of power just to keep the contents from going volatile?
It's a total guess, but if anything it's high. Look at it logically for a minute. The Pandora has a 4000mAH 3.7V battery. 5% of that is 200mAh. Divide that by 24 and you have an 8mA at 3.7V, or 11.5mA at 2.5V (which is what DDR SDRAM actually runs at) allowance for RAM refresh.

Does the RAM draw more than 11.5mA in sleep mode? I doubt it, but I can find no reliable information either way. This information is probably still NDA'd, so I don't expect an official answer until the Pandora ships.
 
Last edited by a moderator:
hey

WAIT...

All this (well quite a bit) has already been talked about in THIS THREAD, and this thread also has some simular things. Hmm, well I like what Squidge has to say in those threads :).

I certainly know what Chip thinks about the subject, as they have also mentioned things on the previous thread, hehe.

Anyhows I actually used google to dig up those threads, I find this forums search function rather limiting or whatever.

cyas
 
Last edited by a moderator:
I used a pair of these in my i.MX31 prototype:

http://www.micron.com/products/partdetail?...46H32M16LFCK-75

They are 1.8V just like the Pandora memory module and are probably pretty much the same internally. From the datasheet on that page, I see 0.5mA per chip in self refresh mode. So assuming it scales up the same in the x32 POP version, then it is 1 mA.

Even with other parts on the board powered down in similar fashion, that sounds like two months of standby time on a fresh battery maybe, but that isn't tested of course!
 
MWeston said:
...that sounds like two months of standby time on a fresh battery...
SWEET JUMPING JEHOSPHAPHAT!!!1¡!¡!1¡!ONE¡ELEVEN!!2

two MONTHS?

That's AWESOME!

I might just be waiting as long as it takes now. Freak, dude. Two blankety-blank months! Wow.
 
Last edited by a moderator:
Back
Top