Page 59 of 64 FirstFirst ... 9495455565758596061626364 LastLast
Results 1,161 to 1,180 of 1263
Like Tree3Likes
  1. giggles's Avatar
    Posts
    677 Posts
    Global Posts
    690 Global Posts
    #1161  
    Quote Originally Posted by wiwa View Post
    I put it in bootmisc.sh behind : exit 0 . You may want to add the softswap script there, since it seems to be working for @GenTarkin. I myself didn't do it because it may conflict with compcache settings oneday when I forget about the altering (I'm very forgetful)

    At nexave.de a Pixi user had to doctor his phone after he tried to edit bootmisc.sh --> it's your own responsibility
    Okay im gonna give that a try after church today. thankx for the warning and happy easter to all.
  2. gfdos.sys's Avatar
    Posts
    69 Posts
    Global Posts
    77 Global Posts
    #1162  
    Hey all. Cool stuff going on here!
    This basically sounds like a windows 95/98 program I used to use called softram... but basically If I understand correctly, we are using part of main ram as a ramdrive and using this as a swap partition....

    All that said I want to point out that the statement that the pixi has 190 MB ram is incorrect. The pixi and pixi plus have 256MB of Ram. Either way, I'm pointing out the obvious here: using 160MB ram as swap, leaves 96MB of ram for actually running programs.... so this is super cool, but has already been stated, kind of "rough" since it does limit how many programs can be run at once.... That got me thinking about the pre 2 which has 512MB of ram...

    ----

    Changing topic a little here, I am currently using a sprint franken-pre2 running webOS 2.1.0, with uberkernel -- so these solutions are posible. (the extra .so modules are present but in joplin folder rather than chuck.)
    And I wanted to see how these modifications would work on the pre2...
    The pre2 has 512MB of ram, the swap partition appears to be 106MB....
    I applyed the changes to bootmisc.sh, and have run softswap160MB.sh...
    now my swap area is actually larger than stock for this device.... and I wanted to test the "speed"....

    The Pre2 with 2.1.0 has flash -- although it is slow... I have tried to watch video , and while the sound was stable the video struggled to keep up.... with this solution in place, its dropping much fewer frames and is watchable.

    This is very very cool because, using 160MB for ramz takes main ram 512MB - 160MB = 352MB to run programs... nice.

    Thanks wiwa!!!
  3. #1163  
    Nice, glad to hear the feedback of this method improving the experience for people!
  4. #1164  
    Quote Originally Posted by wiwa View Post
    We added the following lines to bootmisc.sh:

    Code:
    blockdev --setra none /dev/mapper/store-swap
    blockdev --setra none /dev/mapper/store-var
    blockdev --setra none /dev/mapper/store-media
    blockdev --setra none /dev/mapper/store-update
    blockdev --setra none /dev/mapper/store-log
    blockdev --setra none /dev/mapper/store-root
    the command dmsetup info should then return a read ahead value of 0 for all the partitions.

    I observed a slightly better performance when the swap goes above 60MB w/o compcache. The email app is faster when switching through a bunch of (new) emails. Also the overall system recovered somewhat better when it ran into low mem trouble.

    Background: for some obscure reason the device mapper is optimized for RAID configurations, reading ahead a full stripe of 128KB, expecting sequential read I/O. In fact, there is no relevant amount of sequential blocks on our mobile's flash drive, especially not on the swap partition. So the 128 kilobytes contain useless data in almost every case. I assume that the read cache gets flushed by it.
    Oh, so you actually completely disable readahead for our devices? Because, I am running a 2048kb readahead. I thought read ahead would be better for our devices lol! Guess Ill try no read ahead =)


    Oh, another tip for those trying to automate these settings on boot. Instead of putting these directly in the bootmisc.sh file, I actually put the name of the script in bootmisc.sh, right before the exit 0 line at the end of bootmisc.sh. Then I create the script in /media/internal/scriptname.sh and set it to executeable....that way I can easily modify these settings without having to remount the root filesystem as rw every time I want to change something.
  5. #1165  
    Hrm...weird, I tried the disable read ahead and my UI is vastly laggier now after a reboot =(
  6. #1166  
    Quote Originally Posted by gfdos.sys View Post
    ...

    All that said I want to point out that the statement that the pixi has 190 MB ram is incorrect. The pixi and pixi plus have 256MB of Ram.

    ...


    This is very very cool because, using 160MB for ramz takes main ram 512MB - 160MB = 352MB to run programs... nice.
    Hey, thank you for the feedback.

    Pls don't mind me that I cannot agree to your 256MB statement. Despite being physically equipped with 256MB RAM, the Pixi has less available memory than the Pre. You can evaluate the total memory via cat /proc/meminfo . It's only ~190MB :( The Pre has more then 230MB available memory.

    Your equation 512MB - 160MB* = 352MB is not quite exact. Only the compression result of 160MB is stored in physical RAM, plus the uncompressable pages. I don't now the result for a pegged 160MB space, it must be 60-70MB.

    Quote Originally Posted by GenTarkin View Post
    Hrm...weird, I tried the disable read ahead and my UI is vastly laggier now after a reboot =(
    Are you using Compcache? Try how it works with swap usage larger than 70MB w/o compcache. It's for those who don't like compcache. However, I'll try your values of course
    Last edited by wiwa; 04/25/2011 at 01:53 PM.
  7. giggles's Avatar
    Posts
    677 Posts
    Global Posts
    690 Global Posts
    #1167  
    Quote Originally Posted by GenTarkin View Post
    Oh, so you actually completely disable readahead for our devices? Because, I am running a 2048kb readahead. I thought read ahead would be better for our devices lol! Guess Ill try no read ahead =)


    Oh, another tip for those trying to automate these settings on boot. Instead of putting these directly in the bootmisc.sh file, I actually put the name of the script in bootmisc.sh, right before the exit 0 line at the end of bootmisc.sh. Then I create the script in /media/internal/scriptname.sh and set it to executeable....that way I can easily modify these settings without having to remount the root filesystem as rw every time I want to change something.
    Can you provide some screenshots if possible. Just to be clear I rather not null my device and have to doctor.
  8. #1168  
    Quote Originally Posted by wiwa View Post
    Hey, thank you for the feedback.

    Pls don't mind me that I cannot agree to your 256MB statement. Despite being physically equipped with 256MB RAM, the Pixi has less available memory than the Pre. You can evaluate the total memory via cat /proc/meminfo . It's only ~190MB :( The Pre has more then 230MB available memory.

    Your equation 512MB - 160MB* = 352MB is not quite exact. Only the compression result of 160MB is stored in physical RAM, plus the uncompressable pages. I don't now the result for a pegged 160MB space, it must be 60-70MB.



    Are you using Compcache? Try how it works with swap usage larger than 70MB w/o compcache. It's for those who don't like compcache. However, I'll try your values of course
    Oh so only disable read ahead if not using compcache? What if using compcache? keep the higher read ahead or default?
  9. #1169  
    yeah I reenabled read ahead w / a 1024kb value and its awesome! smoothness definitely came back =)
  10. #1170  
    ^ I kept it completely disabled, because it does no harm. Over the last weeks with this setting I gained more trust in the email app in terms of responsivness. That's good so I'll keep 'setra none'
  11. #1171  
    Something I discovered, I think why using ramzswap as the entire swap w/ no disk backing is so fast is because I ran a quick benchmark:

    Using 4kb block size our SSD's speed is around 19-22MB/sec - sequential read
    Using 4kb block size ramzswap is around up to 30MB/sec - sequential read

    That is much faster for a swap file than the slower SSD =)
  12. #1172  
    The difference is less than I would expect. Can the tool measure the writing speed? I bet we'll see a huge difference there.
  13. #1173  
    Quote Originally Posted by GenTarkin View Post
    yeah I reenabled read ahead w / a 1024kb value and its awesome! smoothness definitely came back =)
    for which partitions, all?
  14. #1174  
    Quote Originally Posted by wiwa View Post
    for which partitions, all?
    I injected 1024 into /sys/block/mmcblk0/queue/read_ahead_kb

    Left the dmsetup thingies on default.
  15. #1175  
    Quote Originally Posted by wiwa View Post
    The difference is less than I would expect. Can the tool measure the writing speed? I bet we'll see a huge difference there.
    If we did write speed testing, that would erase the devices =( lol
    I used dd to do the tests and iotop -o to monitor the speeds

    If I do a write of zero's test...well that erases the pages and worse yet erases the SSD. lol!

    I could attempt a read write test, but I am afraid compcache would be updated too quickly and data might get written back to it thats outdated, causing a crash.

    Also, keep in mind, regarding the speed. Its not as fast as the RAM because the CPU is working its **** off at decompressing pages from the compcache, likewise its compressing when writing to it. Our CPU's are not great general purpose processors and therefore very slow at tasks such as compressing / decompressing.
    Last edited by GenTarkin; 04/27/2011 at 11:57 AM.
  16. #1176  
    Quote Originally Posted by GenTarkin View Post
    I injected 1024 into /sys/block/mmcblk0/queue/read_ahead_kb

    Left the dmsetup thingies on default.
    mmcblk0 is device level, dmsetup is RAID level. So we got overlapping reads ahead Applying the blockdev --setra none to any of the devmapper partitions will affect only the RAID level. It does no harm, while setting the mmcblk0 to zero really sucks.

    Setting mmcblk0 to 1MB indeed loads some apps a little faster. But in my feeling in doesn't accelarate the email app. And I find reading ahead 1MB very aggressive. It might be ineffective for fragmented files
  17. #1177  
    Wait, huh? So...there are overlapping read ahead things goin on on our devices? Should I disable the RAID read ahead and just set the mmcblk0 readahead to 1M or leave the RAID on its default?

    Also, wiwa, I was curious ... do you have any theories about where our extra RAM is gone too on the pixi? Like you say we only have about 200M usable and the other 56M is missing...is it simply not there or is linux somehow hiding it? Any way to find out?
  18. #1178  
    what about a sort of shared video memory for some, you know, graphic accelaration that never comes



    Maybe it could be adjusted via a sort of BIOS, but that's far beyond my skills.

    We're getting way off-topic
    Last edited by wiwa; 04/27/2011 at 06:31 PM.
  19.    #1179  
    Pixi RAM is shared with the baseband modem.

    -- Rod
  20. #1180  
    Quote Originally Posted by rwhitby View Post
    Pixi RAM is shared with the baseband modem.

    -- Rod
    Thanks for clarifying this, is there any way to adjust the shared amount or is it hardcoded in a BIOS of sorts like wiwa stated?

Posting Permissions