Paper evaluations

Kiran Tati (ktati@csag.ucsd.edu)
Thu, 04 May 2000 08:06:14 -0700

The Sprite Network Operating System
The goal of the Sprite project is to design an operating system that
exploits the speed network, large caches and multiple processors in a
single machine. It is designed to share as many resources (disk storage
and cpu of all machines) as possible by the users. The file system of
sprite provide name transparency for the applications by using a prefix
tables. Shared memory is provided to fast communication between
processes, in fact they are threads. Idle machine cpu cycles can be used
by any user in the system and these processes are migrated once the
owner started using the system. Kernel is structured as multiple threads
running concurrently rather than single thread as in standard Unix to
support the multiple processors in a single machine. An RPC mechanism is
implemented in the kernel to communicate with other kernel running on
different machines. Client and server both cache the data to exploit the
large physical memories. Cache consistency is achieved by disabling the
cache when the same file is opened at many places. Virtual memory uses
the ordinary files as backing storage for the simplicity, easy
implementation of process migration and dynamic usage of disk storage.
File System and Virtual memory uses a separate cache.

Caching the virtual memory pages is useful because of locality of
reference of the executing process but the same is not true with the
ordinary file so cache can be used to pre fetch the data rather than
storing the current page of file data. The process migration mechanism
depend on the home machine as well as another machine so if either of
the machine fails the process fails. Not only process migration but in
general system is not fault tolerant, it is not scalable toWAN and it is
not secure.

The Distributed V Kernel and its performance for Diskless Workstations
The goal of this paper is to prove that diskless workstations using the
file server for the data storage is a feasible solution and a general
mechanism/protocol for is as good as well tuned protocol for accessing
the remote services (Disk access). An RPC mechanism is designed for this
system and this RPC is the protocol between the client and server of the
filesystem. An extensive performance measurements are taken to prove
that this general protocol/mechanism, RPC, is as good as any well tuned
protocol. But this RPC mechanism is implemented on top of raw either
net/link layer protocol rather than using the some transport layer for
the performance reasons. The processes names/ids are unique in the
system and they are depend on physical address of the machine (Ethernet
card address) to efficiently finding weather a given process that
belongs to same process or different process so that it can use the RPC
or local procedure call.
In the sequential file access section it is assumed that the latency
for each page is same but this is not true because file system cam
arrange the block so that sequential access is faster.

--
Kiran Tati (Graduate Student)            Kiran Tati
UCSD/CSE AP&M 4402                       9450 Gilman Drive
9500, Gilman Drive, Dept. 0114           UCSD#922563
La Jolla, CA, 92093-0114                 La Jolla, CA 92092-2563
Phone:  858 5345486 (Office)             858 552 9291 (Home)
E-Mail: ktati@csag.ucsd.edu

Office Address: AP & M 6426