 |
Description  |
|
|
RELATED INVENTIONS
The present invention is related to the following inventions, filed on even
date herewith, all assigned to the assignee of the present invention:
1. Title: Nested Contexts in a Virtual Signal Machine.
Inventors: Andrew Kun, Frank Kolnick, and Bruce Mansfield.
Ser. No.: 730,903, now abandoned, and Ser. No. 07/270,437, filed 11/07/88
(continuation).
2. Title: Computer System with Data Residency
Transparency and Data Acess Transparency.
Inventors: Bruce Mansfield, Frank Kolnick, and Andrew Kun.
Ser. Nos.: 07/300,687 filed 01/19/89, a continuation of 07/110,614 (now
abandoned) which was a continuation of 730,929 (now abandoned).
3. Title: Method of Inter-Process Communication in a Distributed Data
Processing System
Inventors: Bernard Weisshaar, Frank Kolnick, Andrew Kun, and Bruce
Mansfield.
Ser. No.: 730,892 now U.S. Pat. No. 4,694,396 issued 09/15/87.
4. Title: Logical Ring in a Virtual Single Machine.
Inventors: Andrew Kun, Frank Kolnick, and Bruce Mansfield.
Sers. No.: 730,923, now abandoned, and 183,469, filed 4/15/88 now
abandoned.
5. Title: Virtual Single Machine With Message-Like Hardware Interrupts and
Processor Exceptions.
Inventors: Andrew Kun, Frank Kolnick, and Bruce Mansfield.
Ser. No.: 730,922 now U.S. Pat. No. 4,835,685 issued 05/30/89.
TECHNICAL FIELD
This invention relates generally to digital data processing, and, in
particular, to a network interface module (NIM) in a distributed data
processing system, in which the NIM comprises means for facilitating
message transmission between processes resident on different cells in the
network.
BACKGROUND OF THE INVENTION
The present invention is implemented in a distributed data processing
system--that is, two or more data processing systems which are capable of
functioning independently but which are so coupled as to send and receive
messages to and from one another.
A Local Area Network (LAN) is an example of a distributed data processing
system. A typical LAN comprises a number of autonomous data processing
"cells", each comprising at least a processor and memory. Each cell is
capable of conducting data processing operations independently. In
addition, each cell is coupled (by appropriate means such as a twisted
wire pair, coaxial cable, fiber optic cable, etc.) to a network of other
cells which may be, for example, a loop, star, tree, etc., depending upon
the design considerations.
As mentioned above, the present invention finds utility in such a
distributed data processing system, since there is a significant need in
such a system for a relatively great degree of hardware independence.
Typical distributed data processing systems comprise a large variety of
diverse processors, memories, operator interfaces, printers, and other
peripherals. Thus there is an urgent need to provide an operating system
for such a distributed data processing system, which operating system will
easily accommodate different types of hardware devices without the
necessity of writing and/or rewriting large portions of such operating
system each time a device is added or removed from the system.
A "process", as used within the present invention, is defined as a
self-contained package of data and executable procedures which operate on
that data, comparable to a "task" in other known systems. Within the
present invention a process can be thought of as comparable to a
subroutine in terms of size, complexity, and the way it is used. The
difference between processes and subroutines is that processes can be
created and destroyed dynamically and can execute concurrently with their
creator and other "subroutines".
Within a process, as used in the present invention, the data is totally
private and cannot be accessed from the outside, i.e., by other processes.
Processes can therefore by used to implement "objects", "modules", or
other higher-level data abstractions. Each process executes sequentially.
Concurrency is achieved through multiple processes, possibly executing on
multiple processors.
Every process in the distributed data processing system of the present
invention has a unique identifier (PID) by which it can be referenced. The
PID is assigned by the system when the process is created, and it is used
by the system to physically locate the process.
Every process also has a non-unique, symbolic "name", which is a
variable-length string of characters. In general, the name of a process is
known system-wide. To restrict the scope of names, the present invention
utilizes the concept of a "context".
A "context" is simply a collection of related processes whose names are not
known outside of the context. Contexts partition the name space into
smaller, more manageable subsystems. They also "hide" names, ensuring that
processes contained in them do not unintentionally conflict with those in
other contexts.
A process in one context cannot explicitly communicate with, and does not
know about, processes inside other contexts. All interaction across
context boundaries must be through a "context process", thus providing a
degree of security. The context process often acts as a switchboard for
incoming messages, rerouting them to the appropriate sub-processes in its
context.
A context process behaves like any other process and additionally has the
property that any processes which it creates are known only to itself and
to each other. Creation of the process constitutes definition of a new
context with the same name as the process.
Any process can create context processes. Each new context thus defined is
completely contained inside the context in which it was created and
therefore is shielded from outside reference. This "nesting" allows the
name space to be structured hierarchically to any desired depth.
Conceptually, the highest level in the hierarchy is the system itself,
which encompasses all contexts. Nesting is used in top-down design to
break a system into components or "layers", where each layer is more
detailed than the preceding one. This is analogous to breaking a task down
into subroutines, and in fact many applications which are single tasks on
known systems may translate to multiple processes in nested contexts.
A "message" is a buffer containing data which tells a process what to do
and/or supplies it with information it needs to carry out its operation.
Each message buffer can have a different length (up to 64 kilobytes). By
convention, the first field in the message buffer defines the type of
message (e.g., "read", "print", "status", "event", etc.).
Messages are queued from one process to another by name or PID. Queuing
avoids potential synchronization problems and is used instead of
semaphores, monitors, etc. The sender of a message is free to continue
after the message is sent. When the receiver attempts to get a message, it
will be suspended until one arrives if none are already waiting in its
queue. Optionally, the sender can specify that it wants to wait for a
reply and is suspended until that specific message arrives. Messages from
any other source are not dequeued until after that happens.
Within the distributed data processing system described herein, messages
are the only way for two processes to exchange data. There is no concept
of a "global variable". Shared memory areas are not allowed, other than
through processes which essentially "manage" each area by means of
messages. Messages are also the only form of dynamic memory that the
system handles. A request to allocate memory therefore returns a block of
memory which can be used locally by the process but can also be
transmitted to another process.
Messages provide the mechanism by which hardware transparency is achieved.
A process located anywhere in the system may send a message to any other
process anywhere else in the system (even on another processor and/or
cell) if it knows the process name. This means that processes can be
dynamically distributed across the system at any time to gain optimal
throughput without changing the processes which reference them. Resolution
of destinations is done by searching the process name space.
The context nesting level determines the "scope of reference" when sending
messages between processes by name. From a given process, a message may be
sent to all processes at its own level (i.e., in the same context) and
(optionally) to any arbitrary higher level. The contexts are searched from
the current context upward until a match is found. All processes with the
given name at that level are then sent a copy of the message. A process
may also send a message to itself or to its parent (the context process)
without knowing either name explicitly, permitting multiple instances of a
process to exist in different contexts, with different names.
Sending messages by PID obviates the need for a name search and ignores
context boundaries. This is the most efficient method of communicating.
There is a significant need to be able to provide within a distributed data
processing system the ability to easily add and delete individual cells in
the network(s) without disrupting network operations.
There is also a significant need to be able to communicate easily and
quickly between processes which may be resident either in the same cell or
in different cells, in a manner which is transparent to the user.
BRIEF SUMMARY OF INVENTION
Accordingly, it is an object of the present invention to provide a
distributed data processing system having an improved network interface.
It is also an object of the present invention to provide an improved
network interface which facilitates message transmission between processes
resident on different cells.
It is another bject of the present invention to provide an improved network
interface which facilitates the addition to and deletion from the network
of individual cells without disrupting network operations.
It is yet another object of the present invention to provide an improved
network interface which facilitates the location of processes resident on
different cells.
It is still another object of the present invention to provide an improved
network interface which facilitates the identification of processes
located on any given cell.
These and other objects are achieved in accordance with a preferred
embodiment of the invention by providing a method of communicating between
processes in a distributed data processing system, the system comprising a
plurality of individual cells and at least two processes resident on
different ones of the cells, the method comprising the steps of providing
in at least a first one of the cells a table of non-resident processes;
generating a request by a first process in the first cell to transmit a
message to a second process identified only by name, the second process
not resident on the first cell; looking up the location of the second
process in the table in the first cell; and transmitting the message to
the cell containing the second process.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention is pointed out with particularity in the appended claims.
However, other features of the invention will become more apparent and the
invention will be best understood by referring to the following detailed
description in conjunction with the accompanying drawings in which:
FIG. 1 shows a representational illustration of a single network,
distributed data processing system incorporating the improved data
management system of the present invention.
FIG. 2 shows a block diagram illustrating a multiple-network, distributed
data processing system incorporating the improved data management system
of the present invention.
FIG. 3 shows an architectural model of a data processing system
incorporating the present invention.
FIG. 4 shows the relationship between software contexts and processes as
they relate to the present invention.
FIG. 5 shows the relationship between external events and processes.
FIG. 6 shows how messages may be sent between processes within nested
contexts.
FIG. 7 shows an architectural model of the improved data management system
incorporating the present invention.
FIG. 8 shows an architectural software model of the improved data
management system incorporating the present invention.
FIG. 9 shows the relationship between pictures, views, and windows in the
human interface of a data processing system incorporating the present
invention.
FIG. 10 shows a conceptual view of the different levels of human interface
within a data processing system incorporating the present invention.
FIG. 11 shows a conceptual view of the local area network (LAN), including
several cells, and a representational view of a network interface module
(NIM).
FIG. 12A shows a representation of the NIM's non-resident process name
cache, and FIG. 12B shows a representation of the NIM's resident process
name cache.
FIG. 13A shows the format of a discovery message.
FIG. 13B shows the format of a discovery response message.
FIG. 13C shows the format of an update cache message.
OVERVIEW OF COMPUTER SYSTEM
With reference to FIG. 1, a distributed computer configuration is shown
comprising multiple cells 2-7 (nodes) loosely coupled by a local area
network (LAN) 1. The number of cells which may be connected to the network
is arbitrary and depends upon the user application. Each cell comprises at
least a processor and memory, as will be discussed in greater detail with
reference to FIG. 2 below. In addition, each cell may also include other
units, such as a printer 8, operator display module (ODM) 9, mass memory
module 13, and other I/O device 10.
With reference now to FIG. 2, a multiple-network distributed computer
configuration is shown. A first local area network LAN 1 comprises several
cells 2,4,and 7. LAN 1 is coupled to a second local area network LAN 2 by
means of an Intelligent Communications Module (ICM) 50. The Intelligent
Communications Module provides a link between the LAN and other networks
and/or remote processors (such as programmable controllers).
LAN 2 may comprise several cells (not shown) and may operate under the same
LAN protocol as that of the present invention, or it may operate under any
of several commercially available protocols, such as Ethernet; MAP, the
Manufacturing Automation Protocol of General Motors Corp.; Systems Network
Architecture (SNA) of International Business Machines, Inc.; SECS-II; etc.
Each ICM 50 is programmable for carrying out one of the above-mentioned
specific protocols. In addition, the basic processing module of the cell
itself can be used as an intelligent peripheral controller (IPC) for
specialized devices.
LAN 1 is additionally coupled to a third local area network LAN 3 via ICM
52. A process controller 55 is also coupled to LAN 1 via ICM 54.
A representative cell N (7, FIG. 2) comprises a processor 24 which, in a
preferred embodiment, is a Motorola 68010 processor. Each cell further
includes a read only memory (ROM) 28 and a random access memory (RAM) 26.
In addition, each cell includes a Network Interface Module (NIM) 21, which
connects the cell to the LAN, and a Bus Interface 29, which couples the
cell to additional devices within a cell. While a minimal cell is capable
of supporting two peripheral devices, such as an Operator Display Module
(ODM) 41 and an I/O Module 44, additional devices (including additional
processors, such as processor 27) can be provided within a cell. Other
additional devices may comprise, for example, a printer 42, and a
mass-storage module 43 which supports a hard disk and a back-up device
(floppy disk or streaming tape drive).
The Operator Display Module 41 provides a keyboard and screen to enable an
operator to input information and receive visual information.
While a single cell may comprise all of the above units, in the typical
user application individual cells will normally be dedicated to
specialized functions. For example, one or more mass storage cells may be
set up to function as data base servers. There may also be several
operator consoles and at least one cell for generating hard-copy printed
output. Either these same cells, or separate dedicated cells, may execute
particular application programs.
The system is particularly designed to provide an integrated solution for
factory automation, data acquisition, and other real-time applications. As
such, it includes a full complement of services, such as a graphical
output, windows, menus, icons, dynamic displays, electronic mail, event
recording, and file management. Software development features include
compilers, a window-oriented editor, a debugger, and
performance-monitoring tools.
Local Area Network
The local area network, as depicted in either FIG. 1 or FIG. 2, ties the
entire system together and makes possible the distributed virtual machine
model described below. The LAN provides high throughput, guaranteed
response, reliability, and low entry cost. The LAN is also autonomous, in
the sense that all system and applications software is unaware of its
existence. For example, any Network Interface Module (e.g. NIM 21, FIG. 2)
could be replaced without rewriting any software other than that which
directly drives it.
The LAN interconnection medium may be twisted-pair or coaxial cable. Two
channels (logically, two distinct networks) may be provided for
reliability and for increased throughput.
The LAN architecture is a logical ring, in which an electronic "token" is
constantly passed from cell to cell at high speed. The current holder of
the token may use it to send a "frame" of data or may pass it on to the
next cell in the ring. The NIM only needs to know the logical address and
status of its immediately succeeding neighbor. The NIM's responsibility is
limited to detecting the failure of that neighbor or the inclusion of a
new neighbor. In general, adjustment to failed or newly added cells is
automatic.
The network interface maps directly into the processor's memory. Data
exchange occurs through a dual-ported buffer pool which contains a linked
list of pending "frames". Logical messages, which vary in length, are
broken into fixed-size frames for transmission and are reassembled by the
receiving NIM. Frames are sequence-numbered for this purpose. If a frame
is not acknowledged within a short period of time, it is retransmitted a
number of times before being treated as a failure.
As described above with reference to FIG. 2, the LAN may be connected to
other LAN's operating under the same LAN protocol via socalled
"bridgeways", or it may be connected to other types of LAN's via
"gateways".
Software Model
The computer operating system of the present invention operates upon
processes, messages, and contexts, as such terms are defined herein. Thus
this operating system offers the programmer a hardware abstraction, rather
than a data or control abstraction.
Processes are referenced without regard to their physical location via a
small set of message-passing primitives. As mentioned earlier, every
process has both a unique system-generated identifier and a not
necessarily unique name assigned by the programmer. The identifier
provides quick direct access, while the name has a limited scope and
provides symbolic, indirect access.
With reference to FIG. 3, an architectural model of the present invention
is shown. The bottom, or hardware, layer 63 comprises a number of
processors 71-76, as described above. The processors 71-76 may exist
physically within one or more cells. The top, or software, layer 60
illustrates a number of processes P1-P10 which send messages m1-m6 to each
other. The middle layer 61, labelled "virtual machine", isolates the
hardware from the software, and it allows programs to be written as if
they were going to be executed on a single processor. Conversely, programs
can be distributed across multiple processors without having been
explicitly designed for that purpose.
An important purpose of the virtual machine concept hereindisclosed is to
provide the applications programmer with a simple, consistent model in
which to design his system. This model, as mentioned above, is reduced to
several elemental concepts: processes, messages, and contexts. As a
consequence of this elemental model, hardware peculiarities are made
transparent to the user, and changes in hardware configurations have no
direct effect on the software.
The Virtual Machine
As discussed earlier, a "process" is a self-contained package of data and
executable procedures which operate on that data. The data is totally
private and cannot be accessed by other processes. There is no concept of
shared memory within the present invention. Execution of a process is
strictly sequential. Multiple processes execute concurrently and must be
scheduled by the operating system. The processes can be re-entrant, in
which case only one copy of the code is loaded even if multiple instances
are active.
Every process has a unique "process identifier number" (PID) by which it
can be referenced. The PID is assigned by the system when the process is
created and remains in effect until the process terminates. The PID
assignment contains a randomizing factor which guarantees that the PID
will not be re-used in the near future. The contents of the PID are
irrelevant to the programmer but are used by the virtual machine to
physically locate the process. A PID may be thought of as a "pointer" to a
process.
Every process also has a "name" which is a variable-length string of
characters assigned by the programmer. A name need not be unique, and this
ambiguity may be used to add new services transparently and to aid in
fault-tolerance.
FIG. 4 illustrates that the system-wide name space is partitioned into
distinct subsets by means of "contexts" identified by reference numerals
90-92. A context is simply a collection of related processes whose names
are not known outside of the context. Context 90, for example, contains
processes A, a, a, b, c, d, and e. Context 91 contains processes B, a, b,
c, and f. And context 92 contains processes C, a, c, d, and x.
One particular process in each context, called the "context process", is
known both within the context and within the immediately enclosing one
(referred to as its "parent context"). In the example illustrated in FIG.
4, processes A-C are context processes for contexts 90-92, respectively.
The parent context of context 91 is context 90, and the parent context of
context 92 is context 91. Conceptually, the context process is located on
the boundary of the context and acts as a gate into it.
Processes inside context 92 can reference any processes inside contexts 90
and 91 by name. However, processes in context 91 can only access processes
in context 92 by going through the context process C. Processes in context
90 can only access processes in context 92 by going through context
processes B and C.
The function of the context process is to filter incoming messages and
either reject them or reroute them to other processes in its context.
Contexts may be nested, allowing a hierarchy of abstractions to be
constructed. A context must reside completely on one cell. The entire
system is treated as an all-encompassing context which is always present
and which is the highest level in the hierarchy. In essence, contexts
define localized protection domains and greatly reduce the chances of
unintentional naming conflicts.
If appropriate, a process inside one context can be "connected" to one
inside another context by exchanging PID's, once contact has been
established through one or the other of the context processes. Most
process servers within the present invention function that way. Initial
access is by name. Once the desired function (such as a window or file) is
"opened", the user process and the service communicate directly via PID's.
A "message" is a variable-length buffer (limited only by the processor's
physical memory size) which carries information between processes. A
header, inaccessible to the programmer, contains the destination name and
the sender's PID. By convention, the first field in a message is a
null-terminated string which defines the type of message (e.g., "read",
"status", etc.) Messages are queued to the receiving process when they are
sent. Queuing ensures serial access and is used in preference to
semaphores, monitors, etc.
As mentioned above, messages provide the mechanism by which hardware
transparency is achieved. A process located anywhere in the virtual
machine can send a message to any other process if it knows its name.
Transparency applies with some restrictions across bridgeways (i.e., the
interfaces between LAN's operating under identical network protocols) and,
in general, not at all across gateways (i.e., the interfaces between LAN's
operating under different network protocols) due to performance
degradation. However, they could so operate, depending upon the required
level of performance.
With reference to FIG. 5, the relationship of external events to processes
will now be described. The virtual machine makes devices look like
processes. For example, when an interrupt occurs in an external device
101, the virtual machine kernel 61 queues an interrupt message 103 to a
specific process 104, known as an "external event service process" (EESP),
functioning as the device manager. For efficiency. the message is
pre-allocated once and circulates between the EESP 104 and the kernel. The
message contains just enough information to indicate the occurrence of the
event. The EESP performs all hardware-specific functions related to the
event, such as setting control registers, moving data 105 to a user's
process 106, transmitting "Read" messages from the user process 106, etc.,
and then "releasing" the interrupt.
To become an EESP, a process issues a "connect" primitive specifying the
appropriate device register(s). It must execute a "disconnect" before it
exits. Device-independence is achieved by making the message protocol
between EESP's and applications processes the same wherever possible. In
this manner a wide variety of different types of hardware devices can be
accommodated without having to make extensive software revisions.
Inter-Process Communication
All inter-process communication is via messages. Consequently, most of the
virtual machine primitives are concerned with processing messages. The
virtual machine kernel primitives are the following:
ALLOC - requests allocation of a (message) buffer of a given size.
FREE - requests deallocation of a given message buffer.
PUT - send a message to a given destination (by name or PID).
GET - wait for and dequeue the next incoming message, optionally from a
specific process (by PID).
FORWARD - pass a received message through to another process.
CALL - send a message, then wait for and dequeue the reply.
REPLY - send a message to the originator of a given message.
ANY MSG - returns "true" if the receive queue is not empty, else returns
"false"; optionally, checks if any messages from a specific PID are
queued.
CONNECT - indicates willingness to accept a particular class of external
events.
DISCONNECT - indicates stop accepting external events.
RELEASE - used by EESP's only to indicate completion of processing of a
particular external event message.
To further describe the function of the kernel primitives, ALLOC handles
all memory allocations. It returns a pointer to a buffer which can be used
for local storage within the process or which can be sent to another
process (via PUT, etc.). ALLOC never "fails", but rather waits until
enough money is freed to satisfy the request.
The PUT primitive queues a message to another process. The sending process
resumes execution as soon as the message is queued.
FORWARD is used to quickly reroute a message but maintain information about
the original sender (whereas PUT always makes the sending process the
originator of the message).
REPLY sends a message to the originator of a previously received message,
rather than by name or PID.
CALL essentially implements remote subroutine invocations, causing the
caller to suspend until the receiver executes a REPLY. Subsequently, the
replied message is dequeued out of sequence, immediately upon arrival, and
the caller resumes execution.
The emphasis is on concurrency, so that as many processes as possible are
executed in parallel. Hence neither PUT nor FORWARD waits for the message
to be delivered. Conversely, GET suspends a process until a message
arrives and dequeues it in one operation. The ANY MSG primitive is
provided so that a process may determine whether there is anything of
interest in the queue before committing itself to a GET. The CONNECT,
DISCONNECT, and RELEASE primitives are specifically used to handle
external interrupts.
When a message is sent by name, the destination process must be found in
the name space. The search path is determined by the nesting of the
contexts in which the sending process resides. From a given process, a
message can be sent to all processes in its own context or (optionally) to
those in any higher context. Refer to FIG. 6. The contexts are searched
from the current one upward until a match is found or until the system
context is reached. All processes with the same name in that context are
then queued a copy of the message.
For example, with reference to FIG. 6, assume that in context 141 process y
sends a message to ALL processes by the name x. Process y first searches
within its own context 141 but finds no process x. The process y searches
within the next higher context 131 (its parent context) but again finds no
process x. Then process y searches within the next higher context 110 and
finds a process x, identified by reference numeral 112. Since it is the
only process x in context 110, it is the only recipient of the message
from process y.
If process a in context 131 sends a message to ALL processes by the name x,
it first searches within its own context 131 and, finding no processes x
there, it then searches within context 110 and finds process x.
Assume that process b in context 131 sends a message to ALL processes by
the name A. It would find process A (111) in context 110, as well as
process A (122) which is the context process for context 121.
A process may also send a message to itself or to its context process
without knowing either name explicitly.
The concept of a "logical ring" (analogous to a LAN) allows a message to be
sent to the NEXT process in the system with a given name. The message goes
to exactly one process in the sender's context, if such a process exists.
Otherwise the parent context is searched.
The virtual machine guarantees that each NEXT transmission will reach a
different process and that eventually a transmission will be sent to the
logically "first" process (the one that sent the original message) in the
ring, completing the loop. In other words, all processes with the same
name at the same level can communicate with each other without knowing how
many there are or where they are located. The logical ring is essentially
for distributing services such as a data base. The ordering of processes
in the ring is not predictable.
For example, regarding FIG. 6, if process a (125) in context 121 sends a
message to process a using the NEXT primitive, the search finds a first
process a (124) in the same context 121. Process a (124) is marked as
having received the message, and then process a (124) sends the message on
to the NEXT process a (123) in context 121. Process a (123) is marked as
having received the message, and then it sends the message on to the NEXT
process a, which is the original sender process a (125), which knows not
to send it further on, since it's been marked as having already received
the message.
Sending messages directly by PID obviates the need for a name search and
ignores context boundaries. This is known as the DIRECT mode of
transmission and is the most efficient. For example, process A (11) sends
a message in the DIRECT mode to process y in context 141.
If a process sends a message in the LOCAL transmission mode, it sends it
only to a process having the given name in the sender's own context.
In summary, including the DIRECT transmission mode, there are five
transmission modes which can be used with the PUT, FORWARD, and CALL
primitives:
ALL - to all processes with the given name in the first context which
contains that name, starting with the sender's context and searching
upwards through all parent contexts.
LOCAL - to all processes with the given name in the sender's context only.
NEXT - to the next process with the given name in the same context as the
sender, if any; otherwise it searches upwards through all parent contexts
until the name is found.
LEVEL - sends to "self" (the sending process) or to "context" (the context
process corresponding to the sender's context); "self" cannot be used with
CALL primitive.
DIRECT - sent by PID.
Messages are usually transmitted by queueing a pointer to the buffer
containing the message. A message is only copied when there are multiple
destinations or when the destination is on another cell.
Further description of the ALL and NEXT transmission modes is found below
in the section entitled DETAILED DESCRIPTION OF INVENTION.
Operating System
The operating system of the present invention consists of a kernel, which
implements the primitives described above, plus a set of processes which
provide process creation and termination, time management (set time, set
alarm, etc.) and which perform cell start-up and configuration. Drivers
for devices are also implemented as processes (EESP's), as described
above. This allows both system services and device drivers to be added or
replaced easily. The operating system also supports swapping and paging,
although both are invisible to applications software.
Unlike known distributed computer systems, that of the present invention
does not use a distinct "name server" process to resolve names. Name
searching is confined to the kernel, which has the advantage of being much
faster.
A minimal bootstrap program resides permanently (in ROM) on every cell,
e.g. ROM 28 in c | | |
