4.4 Multiple Back Ends
As mentioned in Section 2, distributed file systems rarely
offer precise POSIX semantics. Rather, they relax those se-
mantics to improve scalability and performance in a large
distributed system. The details of approximation vary
across systems. Consistency is the most common area of
approximation. Protection, locking, and recording of access
times are other common areas.
To allow flexibility in the choice of distributed file sys-
tem, we plan to restructure ISR software so that most of
the code is written to an abstract file system interface. The
binding of this abstract interface to a specific file system
will be encapsulated in a small back-end code component.
This code component will also determine how VMM files
are mapped to files in the distributed file system.
4.5 Incremental Reconstruction
Another way to reduce resume latency is to allow the
VMM to begin execution before the reconstruction of its
input files is complete. Reconstruction can then be over-
lapped with the resumed execution of the VM. If the VMM
accesses a region of a file that is yet to be reconstructed,
the access has to block until reconstruction of that region is
complete. This effectively combines demand reconstruction
with proactivity. Our implementation will use a stackable
file system approach [9] with the reconstruction manager
residing in user space to simplify experimentation. A mod-
ified Coda kernel module [15] with appropriate extensions
will transparently redirect the VMM’s file accesses to the
reconstruction manager.
5 Alternatives to ISR
This technology addresses the issue of user mobil-
ity. During the 1980’s and 1990’s typical computer usage
evolved from a model in which most users interacted with
managed computer systems to a model in which many users
interact with systems that are either (a) administrated by
that user or (b) customized for that user. Many comput-
ing environments have become funadamentally personal.
ISR attempts to provide a mechanism through which a user
may interact with their personalized computing environ-
ment from different physical locations by leveraging VM
technology.
However, at least two alternative technologies also ad-
dress the issue of user mobility. The first alternative
is remote user-interface (UI) technology as employed by
VNC [13] and X-Move [17]. The basic tradeoff between
remote UI technology and VM migration is that remote UI
technology requires that a fairly steady stream of UI-update
information be transmitted over the network, while VM mi-
gration requires a fairly large state transfer prior to resum-
ing the VM but very little subsequent state transfer until
the time of suspend. VM migration provides a clear advan-
tage in scenarios in which the network may become discon-
nected (such as while travelling by air) or in which network
latency may be high (otherwise, the user experience may be
very poor). Remote UI technology provides a clear advan-
tage when the user interacts with the system for a very short
period of time.
The second alternative is distributed file system (DFS)
technology as mentioned in Section 2. Traditionally, users
have approximated ISR by storing session resume informa-
tion in their home directories on a DFS. Perhaps, many
users will find the ideal mobility solution a combination
of ISR and DFS. In this solution, the OS and related files
are stored in a disk image and the user’s home directory is
stored in the DFS.
6Conclusion
The concept of ISR is obvious once it is described. Yet,
to the best of our knowledge, we are the first to identify
its significance and to demonstrate a working implemen-
tation. It is instructive to ask why this is the case. After
all, IBM had a VM product for its mainframes by the early
1970s. Early distributed file systems, such as AFS and NFS,
emerged by the mid-1980s. Why has it taken so long to
compose the two technologies?
We conjecture that three factors are responsible for the
late emergence of ISR. First, VMs were not available on
hardware typically used by mobile computing researchers
until recently. Only in the late 1990’s, with the founding
of VMware, has VM technology become available on this
class of hardware. Second, most academic research has
tended to focus on process migration as the mechanism for
transferring live execution state. Experimental systems such
as Demos, V, Mach and Sprite have successfully demon-
strated process migration. But there has been no significant
growth of a user community dependent on process migra-
tion. Indeed, there has been a persistent suspicion in the re-
search community that process migration may be a solution
in search of a problem! Third, the growing dominance of
the Microsoft Office application suite has forced researchers
to ask how those applications can be supported in a mobile
computing environment. Since they run only on closed-
source Windows operating systems, solutions that require
operating system modifications are not feasible. A middle-
ware approach such as Puppeteer [3, 5] is one solution. Us-
ing VMs, as described in this paper, is another solution.
Our initial prototype shows that ISR can be successfully
implemented on today’s hardware. We plan to improve the
security, efficiency, and portability of this prototype using
6
Proceedings of the Fourth IEEE Workshop on Mobile Computing Systems and Applications (WMCSA’02)
0-7695-1647-5/02 $17.00 © 2002 IEEE
Resume Templates 1210 Document(s)
