[SI-LIST] Re: Fibre channel interconnect margins

  • From: steve weir <weirsi@xxxxxxxxxx>
  • To: hmurray@xxxxxxxxxxxxxxx,si-list@xxxxxxxxxxxxx
  • Date: Sat, 01 Jul 2006 17:09:04 -0700

Hal, the Gaussian model really does apply and eventually your CPU 
does make mistakes.  It tends to happen at such a low rate we all 
blame it on the software or input devices that make mistakes millions 
of times more frequently.  When it comes to CPUs I am not even sure 
where to begin sifting transistor switching errors from other sources 
because the event frequency is so low.

The problem with understanding Gaussian noise and its unbounded 
nature usually boils down to misclassification of uncorrelated 
deterministic noise as Gaussian noise during measurements.  This 
elevates the Gaussian noise calculation by orders of magnitude and 
depresses predicted BERs similarly.  This accounts for a lot of the 
discrepant observations and comments we have seen on this subject.

Just because we can't readily identify the source of noise does not 
make it random.  It is something of a vexing problem because 
identifying all deterministic noise sources particularly low 
repetition rate sources can be really difficult to do.  But if you 
really want to predict the reliability of your system with confidence 
it is something that you need to at least approach.


Regards,


Steve.
At 04:43 PM 7/1/2006, Hal Murray wrote:

> > Because the statistical  theory behind those errors is that random/
> > Gaussian noise is, by  definition, unbounded - errors are a fact of
> > life, even if the error  rate is very low. Eventually you have to an
> > edge that is outside the  jitter spec. A single unrepeatable error out
> > of billions of bits simply  has to be expected.
>
>By that line of reasoning, your CPU would make occasional errors.
>
>I think the Gaussian noise assumption is reasonable on fiber systems.  If the
>error rate is low, it is exponential in signal/noise ratio.
>
>On copper systems, errors are often associated with crosstalk and/or
>reflections.  Those are deterministic, not Gaussian.  Even EMI and power
>supply noise are probably more deterministic than Gaussian.
>
>A well designed CPU is essentially error free because it's reasonable to
>engineer a system with a signal/noise ratio that turns into less than one
>error per age of the universe.
>
>
>Years ago, when I worked with fibers, consensus was that if you could measure
>the error rate it was too high.  The context was long links - telcom.  Their
>engineering may be less conservative these days.
>
>The error budget included quite a bit of reserve for the laser aging and
>additional splices that get added when backhoes find the fibers.  The systems
>were designed to have a low error rate under worst case conditions.  Most
>systems are far from worst case.  That's especially true just after
>installation when you are trying to verify that everything is working
>correctly.  The extra/reserve signal/noise ratio makes the error rate so low
>you can't measure it.
>
>One trick useful for fiber systems is to add artificial attenuation.
>Reducing the signal reduces the signal/noise ratio which increases the error
>rate.  With the right attenuation you can measure the error rate within a
>reasonable amount of time and then calculate the error rate you would get
>without the extra attenuation.  That works well because the noise on fibers
>really is Gaussian.
>
>
>This thread started when somebody awked about margins in a fibre channel
>link.  Since pre-emphasis and equalization were mentioned, I assume it's a
>copper link rather than fiber.
>
>Has anybody tried inserting attenuators in high speed coax links?  Can you
>learn anything interesting that way?
>
>Is there some way to estimate the deterministic vs Gaussian components of the
>system noise?
>
>If nothing else, increasing the error rate should speed up things like tuning
>equalization.
>
>
>--
>The suespammers.org mail server is located in California.  So are all my
>other mailboxes.  Please do not send unsolicited bulk e-mail or unsolicited
>commercial e-mail to my suespammers.org address or any of my other addresses.
>These are my opinions, not necessarily my employer's.  I hate spam.
>
>
>
>------------------------------------------------------------------
>To unsubscribe from si-list:
>si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
>or to administer your membership from a web page, go to:
>http://www.freelists.org/webpage/si-list
>
>For help:
>si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>List FAQ wiki page is located at:
>                 http://si-list.org/wiki/wiki.pl?Si-List_FAQ
>
>List technical documents are available at:
>                 http://www.si-list.org
>
>List archives are viewable at:
>                 http://www.freelists.org/archives/si-list
>or at our remote archives:
>                 http://groups.yahoo.com/group/si-list/messages
>Old (prior to June 6, 2001) list archives are viewable at:
>                 http://www.qsl.net/wb6tpu
>

------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
http://www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List FAQ wiki page is located at:
                http://si-list.org/wiki/wiki.pl?Si-List_FAQ

List technical documents are available at:
                http://www.si-list.org

List archives are viewable at:     
                http://www.freelists.org/archives/si-list
or at our remote archives:
                http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  

Other related posts: