2x80Gb RAID0 or 1x320Gb?

[kungpo]

It's clear that the use of RAID is a hot subject. It is also evident that (mostly) those who use it love it, and those that don't seem to be fearful of it.

Regardless of statistics and mean time between failures and all that stuff, as long as people are aware of the risk (almost none in my view) then RAID0 is a perfectly acceptable way to get a performance improvement over a single disk.

In work, I use a mix of all types of RAID, and yes, when it fails, it can be a pain in the a$$. But it's not the end of the world if you have a decent backup.

And in any case, re-installing Windows is probably a frequent event for many people here, so not likely to be such a biggie, (I do mine at least every 6 months).

My previous machine was a single disk system. The original disk failed after 10 months, and the replacement (different) failed after 11 months. If my cuurent RAID0 array lasts 10 months or longer, then I have been no worse off, and gained some performance at the same time. Is that so bad?

[iranu]

I'm the sort who doesn't over-generalise.

I'm not sure that staement has any bearing.

In your lone, solitary opinion.

Well I'll back his opinion so he is now not alone and rightly so.

Enlighten yourself:
http://en.wikipedia.org/wiki/Misuse_of_statistics

I'll do this one justice with a smilie. Do you really think quoting that is going to persuade anyone? As someone whose very job depends on the understanding of statistics and how they are applied to material properties I'd lose my job if I misused statistics. Yes misuse occurs for one reason or another but misuse is not occuring here only misunderstanding.

I read through that paper and it is an excellent sorce of data for multi disk (>2 disks) users but the data does not apply to a raid 0 system using 2 disks only. The non-independence of disk failures occuring in that paper is due to previous disk failure. It does not say that because you have two or more disks in a system that there is an increased chance of failure of disk A due to the fact that the system also includes disks, B,C,D,E (until a disk fails) . In a 2 disk raid 0 system we are only concerned with the first failure of a single disk. It makes no difference whether the chance of the second disk failing is increased due to the first, the array is already broken. In that system the chance of failure of one of two disks is independant (until one disk fails). If you do have a failure then it is more likely that the second disk will fail once the array is back up and running based on that paper.

If we didn't use statistics in design for my line of work, engineers and designers would have to use a guestimate. Alot of people would die unnecessarily. I'd hate to be the Chief Engineer and sign off all those designs. In the last 3 years I think I have roughly burned about £2 million on testing and the subsequent data analysis. I suppose this was all a waste of time and we should just make it up as we go along, eh? We have to meet strict CAA and FAA procedures in order to prove that what we are going to fly is going to be safe and that includes our statistical methods, which are laid down in our procedures. We have to use statistics to ensure that we do just that. It is vital that we supply as accurate data as possible because this allows lighter designs with built in safe limits regarding fatigue, which will determine component life and service/maintenence intervals.

I got grilled for 5 hours by a CAA bod during an audit of the department I worked for in 2001 when I was still green with regard to stats. It was most unpleasant because not only did this bod write some of the British Standard material specs that we were using but it turned out that he had a hobby in statistical analysis aswell. He knew exactly what weaknesses we would have and went for it. My boss (30 years experience) and I were terrified because if we didn't pass then projects would be set back months. We passed with flying colours and the only recommendation was that we double the number of materials engineers in the department to 4.

I also remember our Chief Stress man waiting patiently for a result that I was calculating. When I finally came up with the answer he looked a little crest fallen because the figure was too low for the design. After calling over my boss and running it through with him he saw that I had quoted the lower 95% confidence limit. At which point he realised that the design was safe and promptly got up, held my boss by the head and gave him a big kiss on the forehead he was that relieved. 30 odd people were stood around bewildered, but it meant that much to the man to get the correct figure because it backed all of the work done upto that point (we can't get away from concurrent engineering).