I look at your numbers and am afraid that I must be missing something... If
the cost of disk space for a copy of the repository is a few thousand
dollars (let's say, 4 GB at $750/GB) then the redundant space costs you a
few thousand dollars per engineer amortized over the life of the disk.
Let's say you depreciate storage over 3 years. Then you're looking at about
$1K per year per engineer for redundant disk storage. Is this really
excessive? Perhaps I'm missing something and my numbers are way off.
For what it's worth, Microsoft has been issuing publicity
(http://www.microsoft.com/presspass/features/2000/02-28w2k.asp,
http://www.microsoft.com/PressPass/features/2000/02-28w2k-b.asp,
http://www.microsoft.com/TechNet/win2000/fileprin.asp#d) about their
solution to a similar problem at the file-system level for Win2K Server,
called Single-Instance Store (sort of a copy-on-write symbolic link with
sparse files and internal reparse-points so that only the modified portion
of the file is duplicated.). Even if one dislikes MS, their ideas on this
subject can be interesting.
I don't know if there is an analogous technology for Linux, but since the
ideas are out there, perhaps implementing something like SIS for a Linux
(or whatever OS you use) file-system would help address your problem and
would be more generally applicable than just for working with version-control.
If these thoughts are too irrelevant to be useful, sorry to waste your time.
Jonathan Gilligan
At 06:51 PM 4/13/2000, you wrote:
>When you're in company mode, you don't use
>cheap disk on each PC because it becomes
>a mess. You use a network file server
>- with expensive certified fiber disk -
>and a tape backup, because you need to have
>a garantee that your downtime will be less
>than a day per year. With that kind of
>tools, the GB costs you something like $750/GB
>
>Now, if your company is doing chip design,
>some of the files are 600MB EACH. The whole
>repository is several GB...
>You definitively don't want to duplicate this.
