Guys,
I got a phone call from a guy who had questions regarding
the reliability issue. He wishes to remain anonymous, but since this was
a topic discussed in Austin,
I’ve chosen to reply to the reflector.
For those who didn’t attend Austin, here is what happened. I gave a
presentation that examined the pros and cons of SS vs. DS, and part of that was
a look at reliability. (It was mostly a verbal argument and the pdf
contains very little of it.) I concluded that there was no significant reliability
advantage of either architecture, and suggested that common sense therefore
favors the simpler system, which in my opinion is SS.
Hugh disputed my reliability analysis with the following
argument: Hugh claimed that DS is more reliable that SS if the endspan
and midspan are in completely separate boxes (which he claims will be the
normal case). He gave two examples to support this claim:
- Hugh sites the fact that the
fans are usually the least reliable component in these systems. And
he pointed out that a fan failure in one box (the endspan for example) would
not affect the other box (midspan).
- The endspan and midspan could
be on completely different facility circuits, or even in separate
buildings. Therefore a DS system would be less susceptible to
facility power outages.
After that I pretty much let the argument go, mainly because
I was tired and I just wanted to finish my presentation and sit down.
While Hugh’s arguments may seem persusive to some, for
the record I must disagree for the following reasons:
- Just because the endspan and
midspan are in separate boxes (or even separate buildings) doesn’t
mean they offer true redundancy, because they are not equal: If the
endspan goes down for any reason, then the whole system is down because
there is no longer any data traffic. Just because the midspan
continues to power the PD does not mean the system is still “up”.
If you can’t make a call on your IP phone then the system is down.
- I don’t buy the fan
example. If a single fan failure can take down an endspan (or
midspan) then that is a bug in the design of that box, and does not
reflect a weakness in the PoE architecture. Suppose you had a power
supply that kept failing because a cap didn’t have a large enough
ripple current rating; would you fix this by adding a second identical power
supply in parallel? Of course not, you’d fix this design bug
by using a better cap. Network equipment makers know very well about
the failure rates of fans, so they design their boxes with multiple fans,
carefully arranged such that the air flow pattern doesn’t change
radically if one fan dies. (Assuming of course that they know what
they’re doing. I can provide a text book reference if anyone
needs it.)
- I don’t buy the separate buildings
example for two reasons:
- Who is actually going to route
all their cat-5 cables from the endspan in Building A to the midspan in
Building B, and then fan out from there to all the PDs? I think we
can all agree that the endspan and midspan will almost always be in the
same closet, if not the same rack. Although they could be on
separate circuit breakers, but not if they’re running off the same
UPS.
- I think we all know that most
PoE systems use UPS backup. The main reason why many customers buy
PoE equipment is to keep their IP phones alive during a blackout so that
people can call 911. The whole facility power issue seems moot when
a UPS is used.
I suppose we could argue about this forever. (What if
you had the endspan and midspan on separate UPS and separate facility breakers?)
But it would be hair splitting. I still don’t see any clear logical
reason why DS is more reliable than SS, or vise versa. Therefore, I stand
by my original conclusion: Neither DS nor SS offers any significant reliability
advantages, and we’re probably better off just selecting the least
complex system.
Of course all this is just about the reliability aspect of
the SS vs. DS debate, and doesn’t include any of the other pros or cons.
We have to consider everything.
Steve