20071023

Whether we need DCD more than symetry

I'm very much in favor of a symmetrical RJ45 wiring schema for console servers, because it could minimize how many adapters I would need to stock. The advantages of a symmetrical pinout include using a simple 'rolled' cable to make a 'null-modem' connection between devices and adapters.

Now, I'm not picking any particular vendors wiring scheme as 'the best'. I'm espousing symmetry as the 'best way'. The problem is, this means having two Signal Ground pins, which usually means that DCD won't have a pin. (On an 8-wire RJ45 interface, you would have two ground leads, two data leads, two hardware flow control leads, and two hardware handshaking leads.)

Remember that today's Console Server has its roots in yesteryears Terminal Servers. The old TS used modems or terminals, to let remote users connect to hosts across the network. The hardware handshaking was used by the terminals to indicate if they were connected. With modems, they could be connected and powered, but it was DCD signals from the modem to the TS which told the TS whether an interface was 'on-line' with a remote user.

But, these days in console service, modems are a pretty rare thing. Still, I'll be the first to recognize that modems could be used to dial into a remote site and try to connect when the main WAN links are down. However, you can configure most modems to toggle the DSR lead to reflect the status of DCD, and you can configure most console servers to recognize the DSR lead as a signal that the device isn't ready. (So, I don't think we really need DCD in console service. I'll leave Dial-up Networking for later article.)

If you look at my Signals Page, in the RJ45 section, you'll notice that there are already a couple of symmetrical pinout schemes. But you can also see the wide variety of combinations that I've found so far. While pin 4 seems to be the most common pin for signal ground, its clear that not everyone uses pin for for the signal ground. That makes it harder to make an RJ45 signal tracer that works in all cases, so there is less incentive for a vendor to take on that project.

So, for the existing symmetrical RJ45 pinouts on my signals page, you'll notice that they use pins 4 and 5 as ground pins. Data pins are on pins 3 and 6. Handshaking is on pins 2 and 7, and flow control are on pins 1 and 8.

If you're asking "why can't we use pin 5 for DCD?", then you may not have thought about the issue deeply. The first problem comes when you use a rolled cable as a null modem, and you now connect two devices signal leads, but not their signal ground pins! You'll be replying on the chassis ground connections to provide the signal reference. Using these console servers in large data centers means that you may be connecting devices on different phases of the AC circuit, or even across two different Power Distrubtion Units (PDUs) or Uninterruptible Power Supplies (UPSs). The result could be burned out serial ports on your console server and the attached devices, or more sever damage to the units.

So, to summarize the vendors I know about today;

Symmetrical, with 2 grounds (9+ of 26):
Cisco, Logical Solutions SCS, Lantronics SCS-1600, Digi PortServer II, Xyplex 1600/1800, iTouch/MRV in-Reach lines, Juniper Networks (J-series devices, and many NetScreen devices), Server Technology CDU consoles, and Sun Microsystems console/TTY ports

Symmetrical, except for 1 ground and DCD (3 of 26):
Digi CM, Digi STS-1610, Opengear CM

Of the 27 console connections I've documented, 15 use pin 4 for ground, and 9 use pin 6 for ground. (Keep that in mind if you are making your own RJ45 signal tracers.) And of the 9 symmetrical pinouts, eight of them use pins 4 and 5 for ground, and one uses pin6 3 and 6.

It seems to me that other vendors (who are not making console servers) have been choosing the Cisco wiring scheme (Juniper, Sun, Server Technology). Lantronix picked the null-modem compliment to the Cisco pinout. Those choices have been good for me, since it simplifies some of the deployments that I work on. It's interesting to note that the Xyplex/iTouch/MRV pinouts are almost identical, except that pins 1 and 8 (the flow control leads) are inverted from the Cisco schema. (So, if you don't care about flow control, you can mix-and-match these devices with the Cisco-schema cousins.)

I'm under the impression that some vendors are considering adopting a different wiring scheme for their future product series. I'd strongly urge that they consider adopting a symmetrical schema, and I'd also strongly encourage them to consider using an existing schema, rather than trying to come up with a unique schema. And I'm always interested in discussing it in person, though an email on the topic is also welcome.

No comments: