No 40G link from an HP 5900 to Ixia and Spirent load modules over a JG328A DAC, SR4 optics fine
Standing up a small traffic-test bench next to the production fabric, and the tester ports will not come up over copper at all.
- HP 5900 40G switch, JC772A
- HP JG328A 40G QSFP+ direct attach cable
- Ixia Xcellon-Multis QSFP 40/10GE load module
- Spirent TestCenter 40G module, tried as a second opinion
- HP JG325B 40G SR4 transceivers with multimode fibre, as a control
With the JG328A between the 5900 and the Ixia port there is no link at either end. The switch does not throw an error, it only warns that the cable is not recognised and leaves the port down:
<HP5900> display transceiver interfaceSame cable and same switch port against the Spirent 40G module: also nothing. Swap in the JG325B SR4 pair with fibre on those exact ports and both testers link straight away and pass traffic.
What I have ruled out already:
- moved the DAC across three different 40G ports on the 5900, identical result on all of them
- reseated both ends and swapped the cable end for end
- proved the tester ports themselves are healthy using the SR4 optics
So is the JG328A itself suspect, or is this the test equipment refusing a vendor-coded cable? Ordering optics just to feed a load module a couple of metres away is not where I wanted this to go.
Also available in: English
العربية
Deutsch
Español
Français
हिन्दी
Bahasa Indonesia
Italiano
日本語
한국어
Nederlands
Polski
Português
Русский
Türkçe
Tiếng Việt
简体中文
Comments 4
This is the test gear, not your cable and not the 5900.
Load modules from both of those vendors read the cable EEPROM and decide whether they know what is plugged in before they bring the port up. If the coding is not something they recognise, the port simply stays down and you get nothing to look at. HP-coded DACs are squarely in that group. The 5900 is far more relaxed about it: it accepts the cable, prints the warning that it is not recognised, and carries on. That difference in temperament is the whole asymmetry you are seeing.
Borrow a generic, non-HP 40G QSFP+ DAC and put it between the same 5900 port and the same Ixia port. It comes up. The matrix that came out of the bench:
And it is not an HP peculiarity. Other vendors' direct attach cables have a poor hit rate against those testers as well. The practical outcome is to keep one generic cable and your SR4 pair as the lab kit for anything that touches a load module, and use the JG328A where it works perfectly well, which is switch to switch. Nobody is going to recode the tester to suit you.
What does display transceiver interface actually print for that port? Does the 5900 read a vendor name and part number off the cable at all, or are the fields blank? And on the Ixia side, does the port report anything for the inserted cable, or is it completely silent?
One more thing: do you have any non-HP 40G DAC in the cupboard you can put between the same two ports? That single test separates a broken cable from an unwelcome one, and it taes two minutes.
Worth adding, because this bites every time the two ends live in different vendor matrices. We had the same argument on a 100G run: on the switch side Aruba lists R0Z25A, JL307A and R0Z26A as the QSFP28 direct attach cables for the CX 10000, while the server end, an HPE 841QSFP28 adapter, 872726-B21, only lists InfiniBand EDR cables, 834973-B21 through B25. One set is Ethernet-rated, the other InfiniBand-rated, and the adapter does not show up in the compute cable matrix at all because it is a dual-personality card.
There was no cable that both ends called supported. The way out was to spec an Ethernet-only adapter that does appear in the matrix, the MCX516A-CCHT, and then choose a DAC that both sides list. Same reasoning as your bench: buy for the fussier end of the link.
Before you order anything, rule out the boring one. On an ExtremeSwitching X870 we had a DAC that would not link and a 40G QSFP+ module that would not link either, both sitting in QSFP28 ports, and neither had anything to do with who made the cable. The ports were still at their default and had to be told what was in them:
After that both came up. That is EXOS rather than Comware and your 5900 ports are native 40G, so it may well not apply, and the fact that SR4 optics link on the very same port argues against a port-config problem in your case. Check it on your own box before you rule it in or out. The other thing people kept saying about EXOS is that it is quite happy with third-party cables, so brand is rarely where the answer is.