Hi Jason,<br/><br/>Thanks very much.<br/><br/>From your experiences, it seems
that both the fiber cables(multimode and single mode) and copper cables work
without configuring the PHY chips.<br/>Now I am testing a 10Gbe PHY(broadcom
BCM8747) expansion board connected to a FPGA board. System lookpack,independent
lane lookpack,PCS/PMD lookpack operations work. But when I connect two SFP+
ports through mellanox's 3m copper cable(the cable works when I just connect
two ports of a NIC and do a test), it didn't link(PMD RX lose lock and even
can't recieve signal for some ports). I haven't tried a fiber cable(I don't
have one so far). So I am trying to figure out if it is the problem of
configuration of PHY chip like PMA/PMD type.<br/><br/>And thank you for sharing
your experiences on switches and it is helpful.<br/><br/>Best
Wishes,<br/><br/>Lin
At 2014-09-12 01:54:15, "Jason Manley" <[email protected]> wrote:
>No experience on my side with PHY config, unfortunately. Fortunately, I've had
>success with just plugging optics modules into the ports. They seem to just
>work. I've used multimode (-SR) and single mode (-LR) without issue (though I
>can't recall the exact make/model at this time; sorry that's not too
>helpful!). I'd suggest asking for a sample/loaner from your manufacturer of
>choice to try it first.
>
>Are you plugging into a switch or directly to a NIC? I've got no experience
>directly into a NIC, but can offer the following if you've got a switch
>partner:
>
>The 3m copper cables do typically have some bit errors in the direction switch
>-> ROACH, mostly, I suspect, because we haven't tuned the PHY's RX pipeline
>from defaults. But if you use good cables, it's not too bad. Out of 128 links
>currently in use, we have 3 that show single bit BER as poor as ~3.7e-6. I
>consider that pretty bad. It might actually be due to bad SFP+ mezzanine cards
>because they're from a very early prototype batch. The other links are all
>better than 1e-12.
>
>Mellanox have a patch for their switches which boosts their TX and this then
>works really well when using their 3m copper cables. Broadcom devices (almost
>all other common 10G switches out there) seem to work fine out-the-box with 3m
>copper cables. Over the last two months, I've used Huawei CE6850; Arista 7050S
>and 7150S; Extreme Summit X770; HP 5930; A-L Omniswitch 6900; Dell Z9500,
>Z9000 and S4810s; Mellanox SX1036 and SX1012 so far. They all work fine. The
>Mellanox cables are cheap, and are pretty good. I've also used 3m cables from
>Arista and Huawei, and 5m breakout units from HP, and 5m SFP-SFP from Arista.
>These also worked ok. For some reason, the Extreme 2m breakout cables showed
>some more errors than other parts, and consistently had problems interfacing
>to HP 5930s, though I'm not yet sure why because they look like nice beefy,
>well-made, cables. We've also tried some 3m "no-name" brands from Digikey,
>which worked too.
>
>If you really want lossless short-range transmission without having to think
>about any of this stuff, I recommend trying to find some AOC cables. We've had
>good success with these on ROACH-1's CX4 ports.
>
>Jason Manley
>CBF Manager
>SKA-SA
>
>Cell: +27 82 662 7726
>Work: +27 21 506 7300
>
>On 12 Sep 2014, at 5:20, shu.lean <[email protected]> wrote:
>
>> Hi all,
>>
>> Do anyone have experience on configuration of 10Gbe PHY chip? for
>> example,SFP+ mezzanine card that plugs into roach2 is used in two different
>> system.
>> One's transmission distance is more than 300m and the cable is made of
>> fiber. Another's transmission distance is only 3m and the cable is made of
>> copper. How to configure the 10Gbe PHY chips in these two system? Shall
>> PMA/PMD be configured to different types? Or default configuration works in
>> both of the two systems?
>>
>> And since there is only one pair of differential line between the pin of PHY
>> chip and SFP+ module and the data rate is as high as 10Gbps, what about the
>> requirement of PCB materials?
>>
>> Thanks!
>>
>> Lin
>>
>>
>
