В Чтв, 16/09/2010 в 12:54 +0200, Dr. H. Nikolaus Schaller пишет: > Am 16.09.2010 um 10:05 schrieb Gennady Kupava: > > > Hi, Nicolaus. > > > > I have few ideas about about performance. Major bottleneck of freerunner > > is amount of memory, so please can you push as much memory to new device > > as possible. This will make device lightning fast. Second idea is > > The OMAP uses the Package-on-Package concept so we can install > different memory modules depending on what we want, what is available > and what it costs. There are chips with RAM only and chips with RAM/NAND > flash to choose from. > > But we will have at least 256 MByte (I think there aren't any smaller chips).
512 would be excellent. only possible problem is increased power consumption, but one may disable one chip if he case about consumption too much. > > > related to memory too, but i am much less sure about it. Can we have > > some lightning fast sram on device? if bootloader will load kernel code > > into it - this may bring huge speed benefit. > > The OMAP CPU is much faster and many interface controllers are also improved. 'much faster' is enemy of 'fast', so if it is possible to make something even better, why not. But please consider my last idea with great care - i am not hw deleloper just after learning memory subsystem i understood that such system may bring really big benefits. > And, we know that fast boot is possible. At least someone has done it for the > Beagle Board: > http://swiftbeagle.googlecode.com/files/beagleboard_project_hui_keji.pdf > They claim that they have achieved 3 seconds from power up to login: on > the serial console. Well, running X11 also needs some time. From my point of view, boot time especially if it is in 1 minute is not really important. More important how fast it will run while usual usage, how easy is boot system to understand and fix, and how much it deviates from desktop systems. I do not want use busybox shell under any conditions, but running all that services to boot up fully-functional system will take much more time than 3 seconds. (authors use uclibc and busybox, hack init scripts like disable log, disable u-boot menu, disable logs, remove everything from kernel) The way authors of paper archive such boot times influence later speed of device. For example, they propose to use XIP, which will cetrainly decrease kernel speed. Other example is that they 'compiling everything with -Os', it may greatly decrease performance in favor of boot time. PS. please notice your mailer producing doublicating mails (see http://lists.openmoko.org/pipermail/community/2010-September/date.html ) Gennady _______________________________________________ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community