lkml.org 
[lkml]   [1997]   [Nov]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Filesize limitation

While this is an execlent statement, there are uses for files that large:
Consider making a file of a standard type, like a multimeda file (say
.mpg, or .au, or .pcm, or whatever) that contains a very long sequence.

True, if you work on files of this size regulary, you should use a
customized fileformat which makes things like random access sane. But, for
convenience sake you should be able to make a 64 bit file.. Even if
accesses to such files require more overhead (such as a double
dereference)....
This is mostly an issue with the interface between apps and the kernel,
hence, it will be difficult to change.

On Mon, 3 Nov 1997, Richard B. Johnson wrote:

> On Mon, 3 Nov 1997, Andre Uratsuka Manoel wrote:
> [SNIPPED]
>
> > I didn't say it correctly the first time. The file I was
> > reported as having problems being created was slightly larger than 2GB.
> > That file is generated every month and every month it gets bigger and
> > bigger. In about 6 months it will probably not fit into 4 GB either.
> >
> [SNIPPED]
> There appears to be something fundamentally wrong with a program
> that uses such a data file.
>
> In the late '60s IBM created a sort-merge procedure, first used to
> sort the Chicago telephone directory. It became known as the "Chicago
> Sort". It ran on an IBM-360 with 4 kilobytes of core-RAM which had
> to contain both the program and the data. It works.
>
> A few weeks ago, I attempted to use the M$Garbage Editor to edit a 130
> kilobyte text file. It reported "out of RAM" and exited. The two stories
> are related.
>
> Until Software starts being written by Software Engineers, who are
> trained in engineering disciplines, we will continue to have data expand
> like gas to fill all available space. If the space isn't big enough,
> the programs will crash.
>
> Given the current tendency to throw RAM and Disk Drives at a problem,
> it is unlikely that even 64 bits will be good enough in the near future.
> This, in spite of the fact that 64 bits exceeds the dynamic range of
> the universe (233 dB +/- 20 dB).
>
> Even my Sparc won't help. An 'int' on the Sparc is 32 bits. Even if
> you find a 64-bit architecture, that doesn't mean that its file-systems
> will support the kind of file sizes that you propose.
>
> The solution is to use files as files. They have names for very good
> reasons. If a "master-file" is as long as you propose, it contains
> too much information. Such a file should contain "keys" which allow
> records existing in other file(s) to be sorted and merged without actually
> having to copy any data. The records in the other files(s) should contain
> the database information.
>
>
> Cheers,
> Dick Johnson
>
> Richard B. Johnson
> Project Engineer
> Analogic Corporation
> Penguin : Linux version 2.1.60 on an i586 machine (66.15 BogoMips).
> Warning : It's hard to remain at the trailing edge of technology.
>


\
 
 \ /
  Last update: 2005-03-22 13:40    [from the cache]
©2003-2011 Jasper Spaans