I echo Matt's expert summary. The ATA beamformer used to use a great many (initially 24) 3m ibob->bee2 links. We found (as soon as we looked) that the links were very unstable with the FPGA XAUI block reporting nearly-continuous link errors. When XAUI cables from separate iBobs were feeding one FPGA, the synchronization between the links was very unstable.
If you're dead-set on 3m copper cables (or stuck without any hope of fiber), there are a few things that can be done to mitigate the errors (But not always eliminate them): 1. It's very important to keep your iBob cool. Originally our iBobs were cooled only by a head-sink and chassis-level fans. We found this to be insufficient, and were able to correlate link errors with heat sink temperature (a pretty fun test to do). We installed thin fans on each heat sink as matt described and that helped a lot. (Thin fans because that's all we had room for). If you can separate the iBobs and cool them even more, that's better. (Warning: Anecdotal and potentially coincidence-based advice ahead) 2. I found that on the (currently 48) ATA beamformer iBobs, XAUI 0 (as labeled on the front panel) is much more stable than XAUI 1 over 3m links. I think these numbers may be reversed in the toolflow - but the stable one is the further from the Ethernet port. 3. For whatever reason, I found FPGA #2 on the BEE2s to be far more sensitive to link errors than any other FPGA. This is universal among the BEE2s that I tested, both for receiving from an iBob and for transmitting across 3m links to another BEE2. Avoid using this FPGA with 3m copper links. Just my two cents, good luck. Billy -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Matt Dexter Sent: Friday, April 03, 2009 4:32 PM To: G Jones Cc: [email protected] Subject: Re: [casper] iBOB to BEE2 XAUI cable length We experience good performance iBob-> BEE2 over many 1m WL Gore IBN6800-1 cables. We also experience good performance iBob->BEE2 over quite a few 1.5m WL Gore IBN6800-1.5 cables. We have experienced no unusual delays in receiving the 1.5m cables we have ordered even though they are non standard. We ran some limited tests with 2m meterIBN6800-2 iBob->BEE2 and they worked better than 3m and almost as good as 1.5m (surprise surprise). 1.5m were long enough for us so we use those. We experience difficulties over 3m IBN6800-3 for iBob->BEE2 and will not use them in this way unless there is an emergency. We use Zarlink fiber optic CX4 cables for 3m and longer. We also have fans, Sunon GM0504PEV1-8GN, in addition to heatsinks on every iBob's FPGA. Matt Dexter On Fri, 3 Apr 2009, G Jones wrote: > Hello, > I know this question comes up a lot, but is a 3 meter XAUI cable usable for > transmitting data from an iBOB to a BEE2? I have ready Suraj's memo about > testing the link integrity versus preemphasis and differential swing, in > which he says packets were always lost eventually. Is this common in other > people's experience? The memo mentions that tests were also performed with 1 > meter cables, but results are not mentioned for those tests. Do 1 meter > cables perform better? In my system, I can make all the cables 1 meter, but > the routing is more difficult. A 1.5 meter cable would be ideal, but is not > standard. As such, I would like to use 3 meter cables, but am concerned > about data integrity. If 1 meter cables are known to be more reliable than 3 > meter cables for this application, I will use them and deal with the cable > routing issues. > Thanks, > Glenn >
