Today a got a huge update for you.
First, I have made a significant improvement to Live OC. When the OC value is raised not only the bus/GPU speed is increased, but also the DRAM frequency. Since the GPU does not have its own dedicated RAM, but uses a chunk of the DRAM this should increase the stability for high OC values and also the performance of the framebuffer access by making sure that the DRAM does not become a bottleneck.
Furthermore I have improved the performance and responsiveness of the flash memory I/O by carefully tweaking several kernel parameters. This should significantly reduce the probability of pending I/O operations causing lags.
As the third major change I have added the Tiny Preemptive RCU which should further increase the overall responsiveness of the device.
I also updated the optimized CRC32 algorithm to the latest version and removed the CFQ I/O scheduler since deadline is the default and that should be used.
I get a lot requests from users if I could add this or that feature to my kernel and I think it is worthwhile that I tell you a little about my kernel design philosophy.
I come from a Linux background, did my share of kernel hacking before Android and in the Linux world before any patch is even considered to be added to the kernel the author has to prove that this new addition actually does bring an improvement. And believe me the other kernel devs can be quite critical. However it is that kind of peer-review that keeps the Linux kernel of being clogged by nonsense and thus acts as a quality control.
In the Android world pretty much every device has its own kernel and custom kernels are quite popular and every developer essentially is working on his own personal kernel tree. While this fragmentation offers a lot a freedom, it also means that there is no peer-review at all. And unfortunately from what I have seen so far, the general practise around Android kernel developers is to simply add every feature they can find. They do not test, they do not compare, they do not benchmark. The result is a whole parade of overblown and unstable kernels with more unnecessary feature than I can count.
With my GLaDOS kernel I follow a strict no-nonsense policy and try to stay as close to the stock kernel as possible without sacrificing too much performance. Or my '99 with 1' design paradigm as I like to call it - 99% of performance of a theoretical optimum kernel with all possible improvements included while only using only 1% of the necessary modifications to the source code. This does not only improve the maintainability, but in the end will result in a both lean and stable kernel with a great performance.
Since so many people were asking for it, I changed my mind and decided to accept donations. You have my word that the collected funds will only be used for buying additional hardware to continue my development and if possible to extend my work to another device preferably of course the Nexus Prime.