IBM Mobile Computing Workshop, 24 Jan 94
Larry Loucks, VP of Software Architecture, IBM
Keynote Address
He described a taxonomy of hardware and software components related to mobile computing.
Dan Duchamp, Associate Prof., Columbia
Replacement of Replicated File Systems
Dan presented his Summer '93 USENIX paper (click
here
for a copy). I am not a
big fan of this work, in part because it uses NFS, and
in part because I find the model unrealistic.
Charlie Kunzinger, IBM
Summary of IETF Mobile IP Working Group
He described the basic entities:
and the basic procedures:
The RFC is almost ready, and implementations are expected
to follow. (I questioned this strange ordering
of tasks, quoting Dave Clark.)
Little thought has been given to security issues related
to Mobile IP, but it is recognized
the the current proposal is highly vulnerable to "denial of service"
attacks, but then, so is most of IP.
Charlie Perkins, IBM
Mobile Networking
John Constant, Novell
He is the product-line manager for mobile computing at Novell.
In his remarks, his emphasis was on reliability and ease-of-use as the
critical and immediate problems to be solved.
He gave a high-level description of Mobile WorkPlace.
(Click
here for
a related article from Network World.) The basic components of
Mobile WorkPlace 1.0 serve the following functions:
He identified areas of continuing interest at Novell:
Novell sees the requirements of a notebook on a LAN to be their
multi-protocol router product, while on a WAN, a notebook requires
their NetWare Connect, remote MHS, and Mobile WorkPlace products.
He indicated that Novell is doing a lot of research on the networking needs
of personal digital assistants.
M. Satyanarayanan, Carnegie-Mellon
Mobile Computing
Satya gave an overview of his work in mobile computing. He began
by observing that access to shared data is the key problem. While
CD-ROM will serve many of these needs, e.g., manuals for field service
maintenance or emergency medical response, there remains a mutable component,
e.g., the service record or medical record.
What makes access to shared data hard are the fundamental constraints of
mobile computers, namely
and the scale and variety of data.
Satya argues for transparency -- mobile systems should
cope with the effects of mobility.
Satya then described recent work in CODA. He points out that
replay optimization is critical to performance. Work currently
in progress includes
Satya points out that some cache misses are worse than others, e.g.,
a miss in the first minute of service is much worse than one after
24 hours -- in the latter case, the user was able to work for a day.
Peter Honeyman, CITI
Preheating a Mobile File Cache
Here are the foils I used.
Carl Tait, IBM
Hoarding vs. Prefetching
Carl makes an analogy with loading a briefcase.
Panel
I forget exactly who was on the panel, as there was no agenda, no
introductory remarks, and most of the discussion was in the
audience. We argued about the utility (or is it futility?)
of the IETF Mobile IP activities, but no other recollections stand out.
Dave Cohn, Notre Dame
Realizing Mobile Computing Personae
Phil Rogaway, IBM
Confidentiality and Authentication Mechanisms for
the Mobile Environment
Goals:
Problems:
You can see what's coming ... that's right, a new, fast cipher,
called SEAL, for software encryption algorithm. SEAL is designed
to be a software efficient stream cipher. It crunches a pass phrase
into a 160-bit key, using a very slow key schedule mechanism. (This part
takes milliseconds, but is done only once per session. Note how this
aspect of SEAL exacerbates dictionary attack.) You then seed SEAL with an
index, and out flows a bit stream that you XOR with your plaintext.
The index
gives you random access to a file -- if a file has been encrypted with
an initial index of 0, and you want the kth bit of the file, you seek to
the kth bit of the ciphertext, present k as the seed (and a proper pass
phrase), and XOR the resulting stream to recover the plaintext.
This
has the random-access advantages of DES ECB or PCBC without their (horrible)
disadvantages. SEAL looks very exciting. (But what's the chance IBM
will let SEAL be used in the public domain?)
SEAL has been measured on some popular processors:
SEAL is 10-30 times the speed of DES.
He also described some ideas for fast message authentication, based on Wegman and Carters'
universal hash functions, but I didn't get it, nor did anyone else, I suspect.
Handouts
See me for handouts passed out at the meeting. I have
honey@citi.umich.edu