Never crashing night(computers)
5936 cr points
Send Message: Send PM GB Post
32 / M
Offline
Posted 5/21/17 , edited 5/21/17
I'm thinking of a computer cloud setup where it never crashes. The cloud setup is that when one computer crashes, the other takes over. Based on my knowledge of distributed computing, can the program be ran within this joint distributed computing network but not on a single computer? It would need its own operating system and such, but when one computer decided to stop computing, it only loses some processing speed, and the operated system is still running among this joint distributed network. If there is an even simpler solution, let me know, thanks

The core operating system needs to be operated by a set number of computers and back up computers, the rest can adds to its computing power

P.S. Btw the night comes from the game Neverwinter Night
P.S. Shifting processing power for the core operating system might be a little tricky
P.S. Sharing hard drive memory seems to be even harder, maybe use RAM instead
3541 cr points
Send Message: Send PM GB Post
M
Offline
Posted 5/21/17 , edited 5/21/17
Not quite sure what you're asking, try restating it in the form of a question. On a related note, what you're talking about sounds a lot like https://en.wikipedia.org/wiki/OnLive and the process they used for providing cloud computing.
I thought it was pretty high tech stuff a few years back, sadly they got bullied into selling by Sony lmao. As far as I can tell cloud computing has it's set of problems as well, such as dependency on stable internet signal and speed.
62379 cr points
Send Message: Send PM GB Post
29 / M
Offline
Posted 5/21/17 , edited 5/21/17
My information is a little dated, but I seem to recall that this was one of the selling points for using map-reduce on Hadoop, and I'm sure you can find similar features on other distributed computing technologies. The idea is, you (the programmer) write a function that can process individual parts of a whole set of data which can be distributed across the nodes in your cluster. Then the head node determines how to distribute that job so that the individual parts can all be processed in parallel on different hardware. If one of the nodes fails, the head can rebalance the job so that the work that was originally allocated to the failed node can be performed by the other nodes. Typically, the head node has a backup or two that can take over if it fails as well.

What's needed is a way to break up computer programs into pieces that can be run in parallel on separate hardware so that these jobs can be distributed across your cloud system and rebalanced if necessary. However, it does require a paradigm shift for programmers. That said, there is still a limit to what kinds of failures the cluster would be able to handle. That is to say, it would really only be able to recover from hardware failures. If the software fails, there may be no point in rerunning that particular process because there could be a flaw in the logic of the program itself and there's no way for the cluster to correct for that. If the whole program crashes, you can restart the process in the hopes that it can continue to run normally after a restart, but that's probably about the best you can do.
23260 cr points
Send Message: Send PM GB Post
The White House
Offline
Posted 5/21/17 , edited 5/21/17
Its possible to make everything redundant but it comes at a price ($$$$$ and speed). Typically when a server crashes everything on it goes down because it keeps costs down. The last saved state on the hard drive might exist but the data for that session is purged.
62379 cr points
Send Message: Send PM GB Post
29 / M
Offline
Posted 5/21/17 , edited 5/21/17

Rujikin wrote:

Its possible to make everything redundant but it comes at a price ($$$$$ and speed). Typically when a server crashes everything on it goes down because it keeps costs down. The last saved state on the hard drive might exist but the data for that session is purged.


Why would redundancy come at the price of speed? There ought to be a fixed cost in terms of processing time spent while copying data to the redunant nodes which can be easily adjusted for by pouring more dollars into hardware so that it doesn't affect the performance of your application. Furthermore, if you're talking about moving your application to a cloud service, it could be that their hardware ends up being significantly faster than yours. Any speed cost you might incur upon implementing redundancy is going to be entirely dependent on that implementation.
5936 cr points
Send Message: Send PM GB Post
32 / M
Offline
Posted 5/21/17 , edited 5/22/17

staphen wrote:

My information is a little dated, but I seem to recall that this was one of the selling points for using map-reduce on Hadoop, and I'm sure you can find similar features on other distributed computing technologies. The idea is, you (the programmer) write a function that can process individual parts of a whole set of data which can be distributed across the nodes in your cluster. Then the head node determines how to distribute that job so that the individual parts can all be processed in parallel on different hardware. If one of the nodes fails, the head can rebalance the job so that the work that was originally allocated to the failed node can be performed by the other nodes. Typically, the head node has a backup or two that can take over if it fails as well.

What's needed is a way to break up computer programs into pieces that can be run in parallel on separate hardware so that these jobs can be distributed across your cloud system and rebalanced if necessary. However, it does require a paradigm shift for programmers. That said, there is still a limit to what kinds of failures the cluster would be able to handle. That is to say, it would really only be able to recover from hardware failures. If the software fails, there may be no point in rerunning that particular process because there could be a flaw in the logic of the program itself and there's no way for the cluster to correct for that. If the whole program crashes, you can restart the process in the hopes that it can continue to run normally after a restart, but that's probably about the best you can do.


Precisely what I am talking about for mass distributed computing. This is mostly for hardware failure, not sure how to bypass the software problem yet :D. You got one consciousness that can be ran as multiple units, maybe it has something to do with timing
23260 cr points
Send Message: Send PM GB Post
The White House
Offline
Posted 5/21/17 , edited 5/22/17

staphen wrote:


Rujikin wrote:

Its possible to make everything redundant but it comes at a price ($$$$$ and speed). Typically when a server crashes everything on it goes down because it keeps costs down. The last saved state on the hard drive might exist but the data for that session is purged.


Why would redundancy come at the price of speed? There ought to be a fixed cost in terms of processing time spent while copying data to the redunant nodes which can be easily adjusted for by pouring more dollars into hardware so that it doesn't affect the performance of your application. Furthermore, if you're talking about moving your application to a cloud service, it could be that their hardware ends up being significantly faster than yours. Any speed cost you might incur upon implementing redundancy is going to be entirely dependent on that implementation.


Like I said. Speed and $$$$. If I have to write the same data to 2+ places it will be slower than writing to 1 place. To write it to 2 places will cost more $$$.

Server hardware is rarely the best. They have an average time limit they want to get out of the hardware then they want to replace it. Doing that makes sure they get the most bang for their buck in terms of hardware and power/cooling.

Also cloud software is horrible with any games. Too much data being moved around. Even the businesses I work with their cloud software SUCKS because everyone underestimates how much their server will be used and will only upgrade once it becomes a major issue.
5936 cr points
Send Message: Send PM GB Post
32 / M
Offline
Posted 5/22/17 , edited 5/22/17
Hmm, with this in mind, is there a way to run a single program like DOOM across multiple computers in a network? They would probably share the same files in a directory.


P.S. I looked into it a bit and found this link http://stackoverflow.com/questions/3197888/any-c-libraries-to-run-a-single-program-on-multiple-pcs-i-e-use-grid-compu.
You must be logged in to post.