evaluation

Qiao XIN (qxin@cs.ucsd.edu)
Thu, 4 May 2000 01:33:07 -0700 (PDT)

QIAO XIN

Evaluation for the Paper: The Distributed V Kernel
and its Performance for diskless Workstation

The distributed V kernel is a message-based kernel that provides
uniform local and network interprocess communication. It is primarily
used in an environment of diskless workstations connected by a
high-speed local network to a set of file servers and implemented with
the use of a general purpose network interprocess communication
facilities.

The basic model of V kernel is of that small processes communication
in the form of messages and data transfer operation for moving larger
amounts of data between processes. Remote messages are reliably
implemented by writing a interkernel packet on the network. Large
amount of data are transfered directly between users' address spaces
without extra copies and thus reduces buffering or queuing in the
kernel. Remote segment access is implemented using ReceiveWithSegment
and ReplyWithSegment primitives requires less packet transmissions.

The performance evaluation part of the paper emphasis on the cost of
network file access. The V kernel following the pattern of using
request-response form of communication rather than streaming, thereby
reduce the number of protocols needed .It is shown that diskless
workstations can access remote files with minimal performance penalty
and the V message facilities can access remote files at comparable
cost to any well-tuned specialized file access protocol.

The conclusion of the paper is it is feasible to build a distributed
system using diskless workstations connected by a high-speed local
network to file servers using the general V kernel IPC and achieve
satisfactory performance. But I think the file server is still a
critical issue but the paper didn't address a lot. Things should be
concerned like adding more servers, programs execution in the file
server and bottleneck of the network if we have more workstations.


Evaluation of the Paper: The Sprite Network Operating System

Sprite is an experimental network operating system. The building of
the system is motivated by three trends in computer technology:
networks,large memories, and multiprocessors. I think the achievement
of this experimental system is the following:

1. Sharing. This includes sharing of all disk storage and I/O devices
in the network by all processes, sharing of physical memory between
processes on the same workstation, and sharing of process power by
implementing process migration.

2. flexibility. This is provided in the form of prefix tables which
allows user-transparent reconfiguration of the file system and in
the form of backing files which allows workstations to share back
storage.

3. performance. This is provided by using special-purpose RPC protocol
and by using large caches of recently used file blocks stored in
the main memory of both clients and servers.

4. transparency. This includes a transparent network file system and
transparent process migration both to the process and the user. I
think the latter is the prominent feature of Sprite system.

What impress me most is the mechanisms provides by Sprite such RPC,
cache design, and process migration. Though some of them are not new
here, they are combined here and high-performance results. But I am
not clear from the paper about the performance of the virtual
memory-file system negotiation. I think this is critical since both
virtual memory and file system play important role in the system.