Stephen,

Thanks for your continued interest.  We _are_ going to address the latency
and Jan has begun working with us precisely to address the issue of
strange disconnections and better reintegration.  In the longer run the VM
size will go down dramatically too, but that is really longer term.  A
clear message in the rvm setup can be incorporated.

Peter 


On Wed, 1 Jul 1998, Steven N. Hirsch wrote:

> 
> 
> On Wed, 1 Jul 1998 jaharkes_at_cs.cmu.edu wrote:
> 
> > 
> > > Hi Jan!
> > > 
> > > > - What is the rvm data file size you tried to create.
> > > 
> > > I took one of the choices offered by vice-setup; 90M.
> > >  
> > > > - How much physical memory does your system have.
> > > 
> > > The libc-5.4.44 box has 64M of physical memory.
> > > The glibc box has 32M of physical memory.
> > > 
> > > > - How much swap memory.
> > > 
> > > The libc5 box has 20M of swap.
> > > The glibc box has 60M of swap.
> > > 
> > 
> > An in-memory copy is made of the RVM data file. So you initially might like
> > to try a smaller rvm size (44M for the libc5 box, 22M for the glibc box).
> 
> Ok, after your first message I put 2+2 together and realized that I was
> simply short on VM.  After doubling the size of the swap partition on the
> glibc box, it worked as it should.  The installation guide MUST explain
> clearly explain the VM constraint, BTW.  The need for VM >= data partition
> size is by no means intuitive.
> 
> At any rate, I'm up and running over 100Mb ethernet.  The transaction
> latency for creating/removing individual files is still quite high - much
> more noticable than for NFS over the same link.  No problems to this
> point, but a full kernel compile usually breaks things quickly <g>.
> 
> Thanks for the hints!
> 
> Steve
> 
>