A distributed-parallel processing system joins multiple computers
(or multiple CPU's) together into one system (or one machine)
to achieve higher performance than a single computer (or a single
In such a system, the number of computers (or CPU's) is changed
according to performance requirements.
A distributed-parallel processing system is called `scalable'
if the system is able to cope with the change of the number
of computers (or CPU's), i.e. the change of the scale.
There are two types of scalable system; one requires applications
to be rewritten whenever the scale is changed, and the other
just needs simple recompilation or reconfiguration.
SSS-PC is a scalable system of the
- workstation cluster
A workstation cluster is a set of workstations connected with
a high-speed network.
It is also called a `network of workstations (NOW)'.
A server cluster and a PC cluster are varieties of cluster system.
- SSS-PC -- origin
of its name
T. Matsumoto, who is a general clerk and a chief architect and
a chief programmer of SSS-PC, proposed
a synchronization scheme named `Snoopy Spin wait (SS-wait)'
SS-wait was designed to be implemented on a shared memory system
and to be used for synchronization in a multitasking environment
without a drop in performance.
There was an operating system being developed on a shared memory
parallel machine to support SS-wait.
This operating system was called `SS-CORE'.
SSS-PC is a successor to
This new OS expands its aim from a shared memory system to a
workstation cluster or a distributed memory
As the new OS aims at more scalable
(i.e. easier to increase the number of machines) systems, it
is called `Scalable SS-PC', or
`SSS-PC' (pronounced `three ess pc').
- Ethernet / Fast Ethernet
Ethernet is a standardized network technology developed at
It is widely used for LAN (Local Area Network).
Ethernet transmits data at a rate of 10 Mbps (i.e. 10,000,000
bits per second).
Fast Ethernet transmits at a rate of 100 Mbps.
The latest commercial Ethernet is Gigabit Ethernet, which transmits
at a rate of 1 Gbps (i.e. 1,000,000,000 bits per second).
- TCP/IP (Transmission Control Protocol / Internet
IP are communication
protocols for the Internet.
They were designed at the Defense
Advanced Research Projects Agency (DARPA).
They have become the de facto standard both for the Internet
and for LAN.
Although TCP and IP are distinct protocols, they are often used
Therefore they are called `TCP/IP' in a lump.
is a communication protocol based on IP.
IPsec appends security of communication to IP.
TCP/IP and UDP/IP gain security from IPsec
without being aware of it.
IPsec provides encryption, integrity ensurance, and authentication
Encryption protects data from peeping even if the communication
path is monitored.
Encrypted data does not make sense to anyone but the intended
Integrity ensurance means that modifications to data can be
Authentication verifies the identity of the origin of data.
Secure Sockets Layer
(SSL) is another widespread mechanism for secure communication.
SSL is implemented as a part of the user application, while
IPsec is in the system program or the communication hardware.
IPsec implementation is more likely to be high-speed, since
it can cooperate closely with other system programs and the
Multiple computers are connected with a network in a distributed
processing system (or a parallel processing system).
Each computer of such a system is called a `node'.
In a centralized system, one host computer manages the whole
In a distributed-parallel processing system, on the other hand,
every node cooperates with each other to manage the system.
- remote memory access
It facilitates programming for a parallel processing system
if remote memories of the other computers can be accessed freely
as well as a nearby memory of the local computer.
An operation on a remote memory is called a `remote memory
The functionality of a remote memory access is provided either
by hardware or by software.
SSS-PC offers it by software, as
SSS-PC is intended to be used without
A cache is a mechanism to improve effective performance of a
computer by reducing access time for slow storages (e.g. a main
memory and a disk).
When data is read from a slow storage, it is preserved in a
fast storage (i.e. it is cached).
When the same data is to be read again, the preserved data is
used and the original slow storage is not accessed.
Effective performance of a distributed-parallel system is remarkably
improved by caching data on the local node
when remote memory accesses are performed.
But not all copies of data are allowed to be cached simultaneously
because the amount of the local fast storage is limited.
When data is modified and the old data is copied in some cache,
the copied data should be updated or invalidated.
A `cache read miss' means that the cache does not contain the
data to be read.
Some distributed-parallel systems make it possible for their
tasks to move, or migrate, from a node
to another node while the tasks are running.
This is called `task migration'.
In such systems, tasks running on a heavily-loaded node may
migrate to a lightly-loaded node to balance the loads across
The `real' migration is very hard to implement on
Windows or UNIX
because their system call suites for communication are designed
to keep on running on a single node throughout.
SSS-PC employs a virtualized communication
mechanism of MBCF for interprocess
communication to enable the `real' migration.
The `real' migration means that a migrated task can continue
to run even after the original node is shut down.
SSS-PC is equipped with the `real'
migration so that it is possible to increase or decrease nodes
while applications are running.
- official mascot `Manekikko'
`Manekikko' is designed as an official mascot for
SSS-PC, as well as the
and the BSD
Ms. Hiroko Satoh @
us for using a Japanesque lucky cat `manekineko' as a motif.
Ms. Miki Umezawa @ Tama Art
University designed the character under the guidance of
Ms. Fusako Kusunoki.
To SSS-PC Home Page.
© 1998-2003 SSS-PC Project Team.