Coda was originally written on top of Mach 2.6 which did indeed have kernel
threads. However, early experiments showed the kernel threads to have
serious overhead (Re: an IEEE Trans on Computers paper from April '90).
Kernel threads would be a big win, especially the ability to overlap disk
I/O and other processing (currently things block. We do handle network I/O
properly). 

Peter, have you guys implemented cells? If not, have you played with
running multiple venii (to have different, concurrent,  non-overlapping
name spaces?)

Thanks,

david.

Was Re: coda 4.3.13 on RHL 5.0 + egcs 1.0.1
At 10:21 PM 2/3/98 -0500, you wrote:
>On Tue, 3 Feb 1998, Jim Doyle wrote:
>> You guys are doing user-space threads on Linux ??  I had a quick look at
>> the LWP library in Coda.
>
>Yes, for historical reasons.  The coda code has roots in the mid-80s,
>before the widespread availability of threads on Unix.  There is/was some
>support for kernel threads on Mach in some parts of Coda, I think, but I
>don't know if that code has run recently.
>
>Unfortunately, the Coda code may rely on the non-preemptive nature of the
>user-level thread implementation. 
>
>RVM _is_ supposed to be really thread-safe, and one big win would be to
>let the RVM log compaction run in parallel with the other (nonpreemptive)
>threads.
>
>A good project for someone, perhaps...
>
>Michael
>
>
>