So, to give a little background about why the way some people implemented it is a really bad idea...readahead works like this - when you need a section of data from the disk, the kernel will grab that data, and anticipating you'll also use the next X number of kb, it will also grab that data as well and put it into memory. So, when you're doing something like listening to music, or copying data from an sd card (ie long sequential file reads), having a larger readahead is a good thing, and will speed up the process and make things more efficient. But when you aren't doing long sequential reads, you end up thrashing your data. In other words, if you set the readahead value to, let's say 1024kb on /system, every time you access a file you're reading ahead the data that you need, plus and additional 1024kb, or to the end of the file (wouldn't make much sense to read ahead past the end of a file). If you don't end up using that 1024kb it gets flushed from memory, and you read ahead on some other file by 1024kb. You don't end up using that section of data from readahead, it gets flushed, etc, etc. It's a tremendously stupid waste of resources to read ahead that much when you aren't using it. I mean, there's a reason why some of these things are tunable in the kernel and not set to higher values.
And if you want some serious proof check diskstats. With readahead set to 128kb on /system, I still have less than 10% of reads merged. If you only have 10% of reads merged with a 128kb readahead, why on earth (unless you don't know what you're doing) would you want to increase readahead to 1024kb?! To take this one step farther, with readahead set to 4kb, I still only have about 1/3 of the reads merged.
Isn't there a better way to increase readahead?
Yes. The better way is to use Wu Fengguang's series of patches found here http://lwn.net/Articles/372281/. The end result of these patches is that /system, /cache and /dbdata have readahead values of 4kb, /data and your internal and external sd cards have readahead set to 512kb. If you want to take it a step farther and increase it to 1024kb (or whatever value you happen to like - note that you get to a point where you don't get any more throughput, I wouldn't go beyond 1024kb personally), you can do it manually at
echo XXXX > /sys/devices/platform/s3c-sdhci.0/mmc_host/mmc0/mmc0:0001/block/mmcblk0/queue/read_ahead_kb
echo XXXX > /sys/devices/platform/s3c-sdhci.2/mmc_host/mmc2/mmc2:bf2e/block/mmcblk1/queue/read_ahead_kb
What I do is have scripts set up in Gscript lite to increase and decrease readahead, but I don't even use these all the time. Also, if you don't want to flash kernels just to do this, you can set the readahead value for any drive manually, just like for the sd cards,
echo XX > /sys/devices/virtual/block/stl9/queue/read_ahead_kb
echo XX > /sys/devices/virtual/block/stl10/queue/read_ahead_kb
(no point in increasing readahead on /cache, and really, really, really no point in doing it on bml or the other block devices...lol).
In other news...I promise I'll be back soon. I bought a house partially on a whim, partially to spite my girlfriend, and I've been rather busy tweaking the place I live in instead of my phone. But, I just started sorting through the patches I made to my personal sources and I will hopefully have it done tonight...(I know, I've said that many times before, but this time I'm serious...I think)
edit - as an aside, if you've ever wanted to have your display be at the lowered light setting that it switches to just before the screen automagically shuts off, you can control that as well at
echo 1 > /sys/devices/platform/s3cfb/spi_gpio.3/spi3.0/backlight/s5p_bl/brightness