Clock Synchronization

Clock Synchronization

[Dr. Michael Paus]

[-- Attachment #1: Type: text/plain, Size: 830 bytes --]

Hi,

I have a problem which is not directly Ada related but as there
are a lot of embedded system experts in this forum I thought it
is worth a try.

We are running a distributed application on three PowerPC boards
which are connected via VME-Bus. The application is written in
Ada 95 (GNAT) and we use LynxOS as our RTOS. The problem we are
facing is that we need highly synchronized (~1ms) clocks on
each board. We are sending messages with time-stamps back and forth
and so it is important that each part of the application has the
same understanding of the time. Does anybody have an idea of how
to accomplish this without putting too much extra load on the
system? Does anybody know of a hardware solution for this problem
(a common clock which can be read from all CPU-boards)?

Any hints and proposals are welcome.

Michael

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: Visitenkarte f�r Dr. Michael Paus --]
[-- Type: text/x-vcard; charset=us-ascii; name="paus.vcf", Size: 316 bytes --]

begin:vcard 
n:Paus;Dr. Michael
tel;cell:0177-2238312
tel;fax:+49-(0)711-765-4095
tel;home:+49-(0)711-765-1493
tel;work:+49-(0)711-765-4094
x-mozilla-html:FALSE
url:www.ib-paus.com
org:Ingenieurb�ro Dr. Paus
adr:;;Sch�naicher Str. 3;Stuttgart;;70597;Germany
version:2.1
email;internet:paus@ib-paus.com
end:vcard

Re: Clock Synchronization

[Stanley R. Allen]

"Dr. Michael Paus" wrote:
> 
> 
> We are running a distributed application on three PowerPC boards
> which are connected via VME-Bus. The application is written in
> Ada 95 (GNAT) and we use LynxOS as our RTOS. The problem we are
> facing is that we need highly synchronized (~1ms) clocks on
> each board. We are sending messages with time-stamps back and forth
> and so it is important that each part of the application has the
> same understanding of the time. Does anybody have an idea of how
> to accomplish this without putting too much extra load on the
> system? Does anybody know of a hardware solution for this problem
> (a common clock which can be read from all CPU-boards)?
> 

http://www.truetime.com
http://www.datum.com

These companies provide time cards that can be synchronized using
an external signal.  You'll need to have a central source for the
time to distribute to each machine, probably a GPS receiver that
can produce output in IRIG B format.  The cost can be fairly high
for a complete solution.  Your applications will have to use the
time card's clock instead of the motherboard clock.

Another technique is to use the Network Time Protocol (NTP), which
requires that a daemon be running on each machine (ntpd).  One
node is designated the 'server'; its daemon sends it's concept
of time to the 'clients'.  The daemons in the client will adjust
the local CPU clocks, so you can use the local clock.  The NTP
daemons talk over an ordinary LAN, so there's no need for extra
IRIG B wires.

NTP is said to be good for applications needing about 1 millisecond
resolution.  For approx. 1 microsecond resolution, use the time cards.

-- 
Stanley Allen
mailto:Stanley_R_Allen-NR@raytheon.com

Re: Clock Synchronization

[Marin D. Condic]

Dr. Michael Paus wrote:
> We are running a distributed application on three PowerPC boards
> which are connected via VME-Bus. The application is written in
> Ada 95 (GNAT) and we use LynxOS as our RTOS. The problem we are
> facing is that we need highly synchronized (~1ms) clocks on
> each board. We are sending messages with time-stamps back and forth
> and so it is important that each part of the application has the
> same understanding of the time. Does anybody have an idea of how
> to accomplish this without putting too much extra load on the
> system? Does anybody know of a hardware solution for this problem
> (a common clock which can be read from all CPU-boards)?

Try these guys: http://www.datum.com/

When you have multiple computers that you need to synch up to a high
degree of accuracy, nothing beats a hardware solution. The Datum guys
make network time references that are accurate enough for almost any
need. If you need better accuracy, you might want to consult with:
http://www.nist.gov/ http://www.bldrdoc.gov/timefreq/ Of course, getting
the time across the net is pointless since you can't guarantee TCP/IP
propogation delays. You'd have to go with some sort of radio source and
correct for speed of light.

Software synchronization of the different CPU's clocks could be a
problem. It depends a lot on what you have to communicate with between
the different CPUs. Lots of communication mechanisms are not
deterministic, so it becomes impossible to synchronize with any high
degree of accuracy or confidence.

Hope this helps.

MDC
-- 
======================================================================
Marin David Condic - Quadrus Corporation - http://www.quadruscorp.com/
Send Replies To: m c o n d i c @ q u a d r u s c o r p . c o m
Visit my web site at:  http://www.mcondic.com/

"Some people think programming Windows is like nailing jello to the 
ceiling... easy with the right kind of nails."

    --  Ivor Horton - Beginning Visual C++ 6
======================================================================

Re: Clock Synchronization

[Jeff Creem]

"Dr. Michael Paus" <paus@ib-paus.com> wrote in message
news:396ED195.8DED1C1A@ib-paus.com...
stuff deleted.

>
> We are running a distributed application on three PowerPC boards
> which are connected via VME-Bus. The application is written in
> Ada 95 (GNAT) and we use LynxOS as our RTOS. The problem we are
> facing is that we need highly synchronized (~1ms) clocks on


If you are lucky then perhaps the existing boards make their timer registers
available on the bus (is A16 space sometimes)  ...  pick one as the
master.....

Re: Clock Synchronization

[Ehud Lamm]

I am not an embedded man, and don't know much about the hardware you use.
But I thought I'd mention the clock synchonization is a classic
distributed algorithms problem, with several published and analyzied
algorithms.

Ehud Lamm mslamm@mscc.huji.ac.il
http://purl.oclc.org/NET/ehudlamm <== My home on the web 
Check it out and subscribe to the E-List- for interesting essays and more!

Re: Clock Synchronization

[Marin D. Condic]

Ehud Lamm wrote:
> 
> I am not an embedded man, and don't know much about the hardware you use.
> But I thought I'd mention the clock synchonization is a classic
> distributed algorithms problem, with several published and analyzied
> algorithms.
> 
You wouldn't happen to know of any web sites where such
algorithms/papers might be published?

MDC
-- 
======================================================================
Marin David Condic - Quadrus Corporation - http://www.quadruscorp.com/
Send Replies To: m c o n d i c @ q u a d r u s c o r p . c o m
Visit my web site at:  http://www.mcondic.com/

"Some people think programming Windows is like nailing jello to the 
ceiling... easy with the right kind of nails."

    --  Ivor Horton - Beginning Visual C++ 6
======================================================================

Re: Clock Synchronization

[Ehud Lamm]

On Fri, 14 Jul 2000, Marin D. Condic wrote:

|Ehud Lamm wrote:
|> 
|> I am not an embedded man, and don't know much about the hardware you use.
|> But I thought I'd mention the clock synchonization is a classic
|> distributed algorithms problem, with several published and analyzied
|> algorithms.
|> 
|You wouldn't happen to know of any web sites where such
|algorithms/papers might be published?
|

I don't have any good links handy.
But type "clock synchronization" in this serach engine 
 http://liinwww.ira.uka.de/bibliography/index.html#search

and you'll find plenty of papers. The references are to offline media, but
I guess at least some of the papers are available from the authoers
webpages.

Also check this book: http://www.cs.technion.ac.il/~hagit/DC/
It is quite a tome, but worth a look. 5

Last place: Joe Halpern's publication page
(http://www.cs.cornell.edu/home/halpern/abstract.html) has some papers in
PDF/PS format. Just search for "clock"...

HTH.

Ehud Lamm mslamm@mscc.huji.ac.il
http://purl.oclc.org/NET/ehudlamm <== My home on the web 
Check it out and subscribe to the E-List- for interesting essays and more!

Re: Clock Synchronization

[Marin D. Condic]

Ehud Lamm wrote:
> I don't have any good links handy.
> But type "clock synchronization" in this serach engine
>  http://liinwww.ira.uka.de/bibliography/index.html#search
> 
> and you'll find plenty of papers. The references are to offline media, but
> I guess at least some of the papers are available from the authoers
> webpages.
> 
> Also check this book: http://www.cs.technion.ac.il/~hagit/DC/
> It is quite a tome, but worth a look. 5
> 
> Last place: Joe Halpern's publication page
> (http://www.cs.cornell.edu/home/halpern/abstract.html) has some papers in
> PDF/PS format. Just search for "clock"...
> 
Thanks for what you had. I'm sure it will be helpful. I've never found
much published work that dealt with synchronizing computers the ways
I've seen it done in various applications. Especially in applications
where you don't actually have a time-of-day clock, but rather just have
interrupts at regular intervals. I keep hoping I'll trip across some
published work on it. Just never had enough pressing need to do a
thorough search.

MDC
-- 
======================================================================
Marin David Condic - Quadrus Corporation - http://www.quadruscorp.com/
Send Replies To: m c o n d i c @ q u a d r u s c o r p . c o m
Visit my web site at:  http://www.mcondic.com/

"Some people think programming Windows is like nailing jello to the 
ceiling... easy with the right kind of nails."

    --  Ivor Horton - Beginning Visual C++ 6
======================================================================