> The recent discussion about hinting and speculative execution
> brought to mind something I've long wondered about: Why is
> there no way to specify on a file open that you will be reading
> an entire file sequentially? If the OS knew about the caller's
> intentions at open time, it could check the available memory and
> the file's size, and if the ratio were high enough, it could read
> the whole file into a buffer, or even run a background thread to
> do read aheads.
>
> This would only apply to read-only access of existing files, obviously,
> and lacking enough memory, the OS would just ignore the flag an
> do a normal open. Given how often software processes a file in
> exactly this way, it seems like an opportunity to let the programmer
> help the OS perform an optimization.
>
> I've tried to find an example of this being tried in an OS, but come up
> empty. It's such a simple idea that I have to believe it's been looked at
> before.
I'm unclear on the specific flags in specific OSs but I can say for
certain that such things do exist in several research OSs (and I think
in a few production OSs MVS, Solaris and DU? -- I could be off on
these). In regards to the research end of things check out these
projects:
http://www.cs.cmu.edu/~fwc/
http://www.cse.ucsc.edu/~tara/html/thesis.html
http://www.pdl.cs.cmu.edu/TIP/TIP.html
http://www.dcs.uky.edu/~griff/papers/mybib.html (The stuff with Randy
Appleton)
and after all that I can at least tout a little of my research.
http://csl.cse.ucsc.edu/projects/predictive-caching/
and http://www.cse.ucsc.edu/~tmk
The summary of my recent work (HotOS VII) showed that just by tracking
what was opened after the last access to each file, we can correctly tell
you 72% of the time which file will be opened next. I'm currently
implementing this in Linux and at the last Usenix presented a wips
that showed preliminary results (reducing read times by 2/3).
--tmk
----------------------------------------------------------------------- Tom M. Kroeger Pray for wind Graduate Student, UC Santa Cruz \ Pray for waves e-mail: tmk@cs.ucsc.edu |\ and Pray it's your day off! http://www.cse.ucsc.edu/~tmk |~\ (831) 459-4458 |__\ (831) 426-9055 home ,----+--
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/