[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Translucent storage: design, pros, and cons

From: Marcus Brinkmann
Subject: Re: Translucent storage: design, pros, and cons
Date: Thu, 11 Jan 2007 12:48:11 +0100
User-agent: Wanderlust/2.14.0 (Africa) SEMI/1.14.6 (Maruoka) FLIM/1.14.7 (Sanjō) APEL/10.6 Emacs/21.4 (i486-pc-linux-gnu) MULE/5.0 (SAKAKI)


there are aspects of the goals and the design where I am fairly
confident, and aspects where I am less confident, increasingly so as I
learn more about operating system design, from you and other people.

One thing that causes me problems in these discussions between you and
me is that I often don't know if you are talking about just the kernel
or also the user space system when refering to Coyotos.  The space
bank in particular is logically outside the kernel and part of the
user space operating system.  But there seem to be some complicated
interactions between the kernel design and the way memory is organized
in the space bank that I likely do not fully appreciate and which seem
to conflate the issues.

My confusion in other areas of the design space is also a reflection
of the confusion in the research community on these issues.  One could
also call the confusion an opportunity to make choices.  Capability
MAP vs COPY, membranes (or not), active/passive objects etc are merely
some examples for this.  There are also unexpected pitfalls.  Take for
instance the IPC reliability discussion, which ended with agreement
across the board that timeouts or watchdogs are probably the only way
to deal with communication failures sensibly.  This made me aware of
the problem that we might not give sufficient attention to timing
issues in the system.  Hermann Haertig told me at EuroSys that he
thinks it is a mistake to build a system nowadays without
consideration to timing.  This raised an internal alarm flag for me,
which is still raised, waving along until I decide to spend some
resources on the issue.

Which brings me to the next point: I have hardly spent any thoughts on
operating system design in the last half year, and won't spend much
thought on it for the next three months for certain.  I am wrapping up
my math studies in a quite unrelated field (algebra and cryptography).
I am sorry that this leaves you waiting (as well as many other
people), but it is how it is.  I am also emotionally exhausted from
the discussions that happened, and need some time to sit back and
refocus to be able to take a more constructive approach.  I recognize
your translucent space bank as a step in that direction, and one that
is quite appropriate in the context of the Coyotos system design, but
for me this is the second step before the first.  I want to think
about these issues top-down.

Jonathan, you can argue very convincingly from your position, exactly
because you have interwoven many aspects of the design into a single
coherent picture.  All I ask for is to be able to weave my own
picture, and let's not preclude the result.  I have made the mistake
to commit prematurely to a system design once before, disappointing
many people.  Consider me a burnt child.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]