Page 40 of 55 FirstFirst ... 30353637383940414243444550 ... LastLast
Results 781 to 800 of 1081
Like Tree13Likes
  1. #781  
    I've been following this thread for some time now, and I can't wait to take a shot at this This also gives me a reason to re-doctor my phone so I can also have Flash support.
  2. #782  
    Quote Originally Posted by rmausser View Post
    These settings are good, however, I am still finding an issue:

    No matter what we set the swap size to, its obvious that once it hits about 100mb in size, the kernel freaks out and desperately tries to shrink it.

    Once my swap hits 100mb, things still tend to get 'locky' and even once in a while I will get a reboot.

    There must be somewhere telling the kernel 'swap capacity = 100mb'

    We need to find out how to change this to say, 450mb or else our 500mb swap files are useless.

    -- Sent from my Palm Pre using Forums
    Try increasing dirty_ratio to 90. It might be because you have it set so low, the kernel keeps trying to clean vm even though Luna has not maxed out. I was getting lock-ups until I pushed it to 90. Also, I don't know if you 2.1 guys have the swaphack.sh script in the init.d folder, but my 1.4.5 Pixi has it, and it specifies a 100MB swap. You might want to look there to see if that is getting run on start-up and confusing the system.

    Also, I'll ask again, can we confirm that the Luna settings under CPUshare are in MB? I do after all see the one setting with a value of 1024, which leads me to question the units of the values. Also, sorry for being naggy about it, but can anyone give me possible reasons why my ramzswap partition never fills up to 32MB, as I have it set? It seems to max at at about 1/3 of capacity always.
  3. #783  
    Quote Originally Posted by bassman97 View Post
    Try increasing dirty_ratio to 90. It might be because you have it set so low, the kernel keeps trying to clean vm even though Luna has not maxed out. I was getting lock-ups until I pushed it to 90. Also, I don't know if you 2.1 guys have the swaphack.sh script in the init.d folder, but my 1.4.5 Pixi has it, and it specifies a 100MB swap. You might want to look there to see if that is getting run on start-up and confusing the system.

    Also, I'll ask again, can we confirm that the Luna settings under CPUshare are in MB? I do after all see the one setting with a value of 1024, which leads me to question the units of the values. Also, sorry for being naggy about it, but can anyone give me possible reasons why my ramzswap partition never fills up to 32MB, as I have it set? It seems to max at at about 1/3 of capacity always.
    I will try it but I have reservations on dirty ratio.

    My understanding of dirty data is disk cache that the app is using to be reserved in ram for quick access.

    It was designed for spinning hard drives, which have a terribly slow access rate.

    With an ssd like in our Pre's I'd rather have the app access the data from the ssd which has quick read access and use the limited ram we have for application resources.

    I will look into swaphack today, thanks for this info.

    Make sure in compcache you set Ur compcache settings as 'disksize' and not 'memlimit', memlimit does not work correctly with swap (as opposed to backing store), which is a shame as it dynamically resizes compcache, something quite useful.

    Ps: I'm starting to find compcache a bit useless. With a spinning hard disk I could see its potential, but with an access rate of around 20mb/s with our swap partition on solid state memory, and 30mb/s for compcache in ram, you might as well reserve that for real ram and increase swap size on disk.

    -- Sent from my Palm Pre using Forums
  4. #784  
    Just modified swaphack.sh in /etc/init.d

    Phone is flying.

    ONLY IF You HAVE 500Mb swap!!!!!!!! Change size to:

    Code:
    SIZE=400663296
    Regardless of swap comment this out.

    Code:
    #echo 6000 > /proc/sys/vm/dirty_expire_centisecs
    Its like a triple shot of espresso for your Pre.

    -- Sent from my Palm Pre using Forums
  5. #785  
    Well, I am using disksize, but for whatever reason it doesn't fill up. Should I play with the priority level maybe? Maybe something else in the Pixi is interfering, that I didn't find yet.

    For dirty_ratio, this was my understanding:

    dirty_ratio | LinuxInsight

    According to them, it's the point when writing out begins. At least on the Pixi, the writing out caused my phone to lock-out, as the low value would basically be hit always, causing the write-out to suck up processing power.
  6. #786  
    Quote Originally Posted by rmausser View Post
    Just modified swaphack.sh in /etc/init.d

    Phone is flying.

    ONLY IF You HAVE 500Mb swap!!!!!!!! Change size to:

    Code:
    SIZE=400663296
    Regardless of swap comment this out.

    Code:
    #echo 6000 > /proc/sys/vm/dirty_expire_centisecs
    Its like a triple shot of espresso for your Pre.

    -- Sent from my Palm Pre using Forums
    if i used your patch in the make file i am using 500mb correct?
  7. #787  
    Quote Originally Posted by bassman97 View Post
    Well, I am using disksize, but for whatever reason it doesn't fill up. Should I play with the priority level maybe? Maybe something else in the Pixi is interfering, that I didn't find yet.

    For dirty_ratio, this was my understanding:

    dirty_ratio | LinuxInsight

    According to them, it's the point when writing out begins. At least on the Pixi, the writing out caused my phone to lock-out, as the low value would basically be hit always, causing the write-out to suck up processing power.
    Well there's the difference.

    Most Pre- users running 2.1 have their cpus clocked at 1ghz.

    Plus the pre cpu runs circles around the pixi as it is.

    So I think we can afford the loss of some cycles here and there if it means more ram.

    Still gonna try it at 90 tho.



    -- Sent from my Palm Pre using Forums
  8. #788  
    Quote Originally Posted by graffix31 View Post
    if i used your patch in the make file i am using 500mb correct?
    I have no such patch that changes Ur swap partition to 500mb.

    Infact that is probably the one thing I won't ever make a patch for. Its too risky to be automated.

    -- Sent from my Palm Pre using Forums
  9. #789  
    your right it was carrel. he gave me a patch to the make file when i made my dr. that increased the swap partition. swapon -s says 499704 as my partition size. what number do i need to use in my swaphack file? thanks for the help.
  10. #790  
    I dont really know haha, I just changed the 1 to a 4, hoping it will be 4 times as large as 100mb.

    so I made it this:

    Code:
    SIZE=400663296
    im not even certain if this works.

    It seems to, im at 125mb swap and no lag.
  11. #791  
    Thanks. Running it now. So far so good.

    -- Sent from my Palm Pre using Forums
  12. #792  
    ANNNNNNND we have a game changer here:

    Code:
    vm.swappiness = 90
    vm.vfs_cache_pressure = 50
    vm.dirty_expire_centisecs = 1200
    vm.dirty_writeback_centisecs = 500
    vm.dirty_background_ratio = 50
    vm.dirty_ratio = 90
    vm.min_free_kbytes = 1024
    vm.max_map_count = 131072
    vm.overcommit_memory = 1
    vm.overcommit_ratio = 70
    vm.zone_reclaim_mode = 2
    kernel.shmax = 268435456
    kernel.shmall = 16777216
    sched_compat_yield = 1
    vm.page-cluster = 6
    vm.lowmem_reserve_ratio = 1024
    Dont think, dont ask, just do it and thank me later....
  13. #793  
    just switched to it. hopefully the thanks will be coming. :-)
  14.    #794  
    ok I'l give it a shot. Why does this change the game?
  15. #795  
    will there be a patch for this?
  16. #796  
    Quote Originally Posted by rmausser View Post
    ANNNNNNND we have a game changer here:

    Code:
    vm.swappiness = 90
    vm.vfs_cache_pressure = 50
    vm.dirty_expire_centisecs = 1200
    vm.dirty_writeback_centisecs = 500
    vm.dirty_background_ratio = 50
    vm.dirty_ratio = 90
    vm.min_free_kbytes = 1024
    vm.max_map_count = 131072
    vm.overcommit_memory = 1
    vm.overcommit_ratio = 70
    vm.zone_reclaim_mode = 2
    kernel.shmax = 268435456
    kernel.shmall = 16777216
    sched_compat_yield = 1
    vm.page-cluster = 6
    vm.lowmem_reserve_ratio = 1024
    Dont think, dont ask, just do it and thank me later....
    Any reason why lowering the cache pressure helps now or did you mean to write 500 instead of 50?

    I also found an answer to my ramzswap question. According to the google code page for the project, when you specify disksize, that is the amount of disk reserved for the space. However, only uncompressed data equal to that size gets stored. So in my case, with a 32MB partition, 32MB of data was stored, but only 10MB of the partition was used, as that 10MB equaled 32MB uncompressed. That is why the compcache developers use a backing disk, as using the compcache settings in Govnah allows the full 32MB to be used prior to using regular swap. With the mods to the system conf files and regular compcahe enabled, things are quite rosey or me right now.

    For me, systcl looks like this for me:

    Code:
    vm.swappiness = 90
    vm.vfs_cache_pressure = 300
    vm.dirty_expire_centisecs = 6000
    vm.dirty_writeback_centisecs = 18000
    vm.dirty_background_ratio = 5
    vm.dirty_ratio = 90
    vm.min_free_kbytes = 3072 (I uncommented the bit in the file that specified Luna to overcommit)
    vm.overcommit_memory = 0
    vm.page-cluster = 4
    vm.oom_kill_allocating_task = 1
    And now with compcache filling up, 32MB may be too big, so I may shrink it to 20 or 24 MB. I'll have to see as days go by. Same thing goes for the Luna settings, to see if those changes keep Luna smaller or enable better clean-up. Are there any docs out there that explain what the settings in the Luna conf file does? But overall, this discussion has made a world of difference to my phone. I have not experienced a lock-up in a while and my only major complaint is that the preference to swap delays my phone answers 1-2 seconds, but that is better than not receiving it.

    I wonder if HP/Palm is reading this. Hopefully they are since this thread is basically free memory optimization info for them.
  17. #797  
    Quote Originally Posted by bassman97 View Post
    That is why the compcache developers use a backing disk
    They've also removed backing swap support in future versions:

    Changelog - compcache - Compressed Caching for Linux

    Because they themselves say it sucks:

    LKML: Nitin Gupta: [PATCH] ramzswap: Remove backing swap support

    Bottom line: No, the devs don't think it's a good idea. They suggest to just use a second swap partition/file nd let Linux "naturally" take care of the work.

    Quote Originally Posted by bassman97 View Post
    I wonder if HP/Palm is reading this. Hopefully they are since this thread is basically free memory optimization info for them.
    They don't need to. They're "making devices" that have enough RAM to begin with so they don't need to play these games anymore.


    M.
  18. #798  
    Quote Originally Posted by Xanadu73 View Post
    They've also removed backing swap support in future versions:

    Changelog - compcache - Compressed Caching for Linux

    Because they themselves say it sucks:

    LKML: Nitin Gupta: [PATCH] ramzswap: Remove backing swap support

    Bottom line: No, the devs don't think it's a good idea. They suggest to just use a second swap partition/file nd let Linux "naturally" take care of the work.



    They don't need to. They're "making devices" that have enough RAM to begin with so they don't need to play these games anymore.


    M.
    Hmm, I must have read old comments then and jumped ship too quick. Looks like I need to find a different way to make sure that the full disksize has 32MB of compressed data in it vs. 32MB of original data.

    But regardless of large memory or not, it is still helpful to optimize to get the best performance possible. Just because my truck has a bigger gas tank now doesn't mean I'm going to ignore the 13 mpg it gets. Similarly, HP/Palm should have the same mentality instead of being lazy to do a google search to arrive here.
  19. #799  
    I've watched this thread also as my pre minus w/ 2.1 needs restarted a lot due to swap lag. Is this close to being in preware? I'd be anxious to try it once its stable Thanks
  20. #800  
    Quote Originally Posted by RyanCell View Post
    I've watched this thread also as my pre minus w/ 2.1 needs restarted a lot due to swap lag. Is this close to being in preware? I'd be anxious to try it once its stable Thanks
    Stability isn't the problem, someone just needs to make a script to add/modify these conf files and parameters in them. The issue now is just finding what is the best possible values are for all the settings, though I have a feeling that there will probably be at least 3 versions of these settings:

    1. For devices w/ 256 MB.
    2. Pixi
    3. For devices with 512 MB

    Let's not forget kernel and software versions also play a role. I am already using a different version in my 1.4.5 Pixi vs. the 2.1 Pre- guys. I wonder if in the case of the vm settings, if Govnah can be modified to include these settings as parameters that it can "govern." This would at least take care of the different hardware/software iterations problem and leave the patching to the Luna conf file.

Posting Permissions