| Title: | FDDI - The Next Generation |
| Moderator: | NETCAD::STEFANI |
| Created: | Thu Apr 27 1989 |
| Last Modified: | Thu Jun 05 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 2259 |
| Total number of notes: | 8590 |
I have a problem at my customer's site with respect to FDDI
or GIGAswitch/FDDI latency I think.
In a nutshell, the customer has Windows-NT boxes running an X-Windows
application from a Alpha 4100 VMS box. They are all connected via a
GIGAswitch/FDDI.
The 4100 is on a GIGAswitch port by itself, and the WIndows-NT boxes
are connected to an FDDI concentrator which is hanging on another
GIGAswitch/FDDI port.
When they eliminate the GIGAswitch, the X-windows application has
IMMEDIATE response to keyboard/mouse events. (In fact, during the
prototype they used a single ETHERNET and it also produced NO
NOTICEABLE latency). When they use the GIGAswitch as described above,
they have perhaps 100-250mS (1/10th to 1/4 of a second) of delay
from mouse-click to screen affect.
I can't seem to determine where this latency is coming from. Customer
is happy right now, and this is a tiny 'annoyance' RIGHT NOW. BUT, I am
trying to avoid a blow-up down the road when they ask us for a
resolution or at least more information. This will happen. This is
being used in a real-time factory-floor application (metal producing).
Can anyone help identify a process to track this latency down?
I have actual design drawing of the network if it will help.
Contact me offline.
Mark
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 2218.1 | some questions | NETCAD::ROLKE | The FDDI Genome Project | Tue Feb 11 1997 10:26 | 12 |
Hi, A simple test is to ping the host from the end node. What is the round trip delay? What other traffic is happening on the FDDI? Do you have other nodes on the ring generating enormous traffic loads (as in a disk farm backup)? What kind of statistics do the FDDI interfaces show? You want to see no errors and few Ring Inits... Chuck | |||||
| 2218.2 | NPSS::MDLYONS | Michael D. Lyons DTN 226-6943 | Tue Feb 11 1997 17:18 | 26 | |
...this sort of thing comes up occasionally in the NPSS::GIGASWITCH
notes conference. See the keyword "performance" for some related
discussions.
Inserting any active device will add latency, although the unicast
latency on the GIGAswitch/FDDI system is rather low (15 microseconds).
Many times, people encounter added latency because frames are being
flooded instead of being sent to just the desired destination port.
This happens because some protocols are one-sided, and the destination
address ages out. Flooding performance is greatly different than unicast
performance since it requires processing by the SCP.
Occasionally, there are misconfiguration issues which only come to
light when the topology is changed, like having multiple ports with the
same MAC address.
I would start by tracing the packet flow, and checking the
forwarding database on the GIGAswitch/FDDI system to see what the
GIGAswitch/FDDI system believes is happening with the MAC addresses in
question.
Also, as .1 suggests, it may be a new contention related problem
which only came about when the topology was changed.
MDL
| |||||