|
2#

楼主 |
发表于 2008-9-11 14:06:03
|
只看该作者
This trick, known as Reed-Solomon error correction, is employed routinely to interpret the data on DVDs, but it has not been used before in the volatile world of private computers on the internet. The first step is to convert the file (which is, regardless of what it represents, simply a long string of ones and zeros) into a mathematical function called a polynomial. This is done by splitting it into 100 fragments, which are smaller binary numbers. It is these numbers that are used to define the polynomial.
One of the characteristics of a polynomial is that a few numbers can nail it down precisely. If a simple polynomial is plotted out on a graph it forms a line. A straight line (the simplest type of polynomial) is defined by any two points on its length. A parabola can be defined by three points. The polynomials that Wuala generates can be defined by 100 points—though, because the polynomials used are not simple ones, these points are complex mathematical constructs, rather than straightforward numbers.
All you have to do now, therefore, is select a suitable number of points from along the polynomial (these need not be the original ones) and convert their values into the appropriate mathematical constructs. Scatter these around the host computers and, when someone wants to look at the file, he need recover only 100 of them to have enough data to reconstruct the file from scratch. To have 100 points available 999,999 times out of a million it turns out that you need to scatter a total of 600 of them around. That is an amount of data equivalent to six versions of the original file, rather than the 100 that would be needed to achieve the same level of reliability if whole files were being stored. Moreover, the system needs the computers linked to it to be available for only 17% of the time, rather than 25%, for this to apply.
Online storage is a growing market, especially for backing up data, where reliability is a big concern. Most commercial online-storage services use centralised servers. Although these are generally reliable, they do sometimes fail. And when they do, the results are embarrassing—as Amazon, an online shopping company, learnt on two occasions this year when the servers for its commercial data-storage system went down for several hours at a time. |
|