Science Forums 2

[alexander]

Can anyone tell me why this website does not have an autosave feature like gmail does?

i just lost 2 pages worth of post because someone clicked the back button.... not really motivated to so that again, it was an excellent post, furious!

Ok for the second time for me, the discussion starts with datastructures...

There are many datastructures in the world of programming, some built into the languages like int , char and bool, others are user defined, some made for speed of operation and virtually unlimited amount of possible data storage, with a sacrifice of memory, like vectors, some for best use of memory but pain of operation, like linked lists and especially queues. Yet there are some datastructures that are optimised for speed of reference and search, those are the ones that we'll be talking about today...


Ever since the first file systems have come out, man, or programmers have been striding to find a perfect structure for speed and efficiency of operation and search. Most of us have heard of GoogleFS with their amazing hashing and searching algorithms with trees. Well, there is a team of people who have developed a file system that when comes out (stable) will be a super crazy fast, far advanced ahead of its time, free and open-source file system that will, I hope dominate the world. Referring to Reiser here.

Reiser is based on a simple concept of a tree data structure. Basically you have a node, the tree, that points to one to ususally 3 nodes or leaves, thus making a tree shape which earned it (the structure the name tree, here's the visual:

yeah kinda like that.

Anyhow, tha main advantage of this is that it is easier to traverse the tree to find something in it, disadvantage is that they are a pain to traverse, and manage as a programmer.

So what is so special about the Reiser that i call revolutionary? well its not just simple trees that i am referring to, the guy created a datastructure that he called a dancing tree. Its crasy even in concept, its a tree that ballances itself to optimize the search though itself... Yeah if you have no idea what i just said, your jaw should have dropped to the floor... Its well known that trees usually dont need to be perfectly ballanced to be easily searchable, well this guy made sure that the trees arent infact ballanced, but infat are built in such a way that will make then easier to iterate through.

The FS also uses something like hashing, where it generates keys for the data and then instead of trying to find some sort of data, the search concentrates on keys, which makes it faster.

So why do i think that Reiser 4 will be a revolutionary acheivement and will be far more advanced then anything we are yet to see? well, to raise the craziness factor to 1 superultragooglogazillion, as if their datastructure wasnt crazy enough, they have created a modulated file system. (your jaw should be a few floors down now and hitting the basement floor) What do i mean by modulated? just that, you will be able, well already are but not too stably, to write modules for your filesystem. Here's an example, say you have a load of backup data stored on tapes and you want to easily reference the files without having to search for them, well, a module that will create a file and point to an exact tape, for its location would allow youto do that, so when you access the file, the FS will ask you to put in certain tape that contains that backup file to retreat it. "Heh" you say, you can make a lib to do that, and there were some programs that let you do that in past. Well, thats how i thought of this example, you see, imagine the speed differece in a library operation vs a module for a file system? yeah i thought a pretty major difference too... also you will be able to do filesystem upgrades with no repartitioning or remaking the fs with modules... (the jaw should start to be digging into the ground)

As if that wasnt ehnough, there are new precotions that this FS will make super nifty and safe. The have developed an algorithm that will either allow you to make a full transaction or not make one at all, without having to copy the data twice! (your jaw should be digging into the ground a few feet under your basement right now)

Anyhow, be weary of using Reiser 4 for now, its really unstable, and still has many bugs, reiser 3 is ok for the moment, but when all the debugging is done, think of the possibilities

anyways, if you are interested, here is th efull description page from the makers of reiser:

http://www.namesys.com/v4/v4.html#trees_nodes_items

and reiser 4 benchmarks:

http://www.namesys.com/benchmarks.html