| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
I accept the utility of supporting 40G servers. My question was really about how to do it, which you touched on in your second paragraph. There could be multiple ways of scaling 8x50G to 8x40G. For examples, one could slow the electrical interface from 25G lanes to 20G lanes, or use a fraction of the available throughput by dumping the excess/unused bits. Neither of these is as simple as making collections of four lanes to form 100G break-outs. I hope some contributions come forward that explore these alternatives, not just to show they are practical, but to show the pros and cons of the alternatives. I am concerned with needing complex high-speed gear box chips in the modules that will dissipate extra power and limit form factor reduction. Paul From: John_DAmbrosia@xxxxxxxx [mailto:John_DAmbrosia@xxxxxxxx] Paul, As demonstrated in the CFI 40G servers are already being deployed. This will grow. One thing I would note though is something that Brad brought up earlier, which I support. The aggregate of higher density support for 40G does not mean that the aggregate itself would necessarily have to add up to 400G. Think 8x50 for example. We could run that at 8 ports of 40G, the aggregate is not 400G, but the breakout function has been supported. John From: Kolesar, Paul [mailto:PKOLESAR@xxxxxxxxxxxxx] I am in favor of an objective for break-out functionality. That said, the challenge of breaking out 40G interfaces from a system based on 25G electrical lanes does not appear trivial. Fortunately this is not the case for 100G break-outs. While the proposed objective language does not mandate 40G break-out, I am concerned that stating it as a possibility could become a distraction, consuming time and resources. Can someone state a case that supports keeping 40G in the proposed objective? Regards, Paul From: John D'Ambrosia [mailto:John_DAmbrosia@xxxxxxxx] Chris, There are a number of ways to enable breakout, which could be via the architecture itself. Also, note that there is no timing related to this. It might apply to a future PMD. And as Mike noted in his response, there is nothing in here that says all PMD variants have to support break out. John From: Chris Cole [mailto:chris.cole@xxxxxxxxxxx] John, Does this objective preclude a duplex SMF (pair) PMD, for example 8x50G or 4x100G WDM? Chris From: John D'Ambrosia [mailto:John_DAmbrosia@xxxxxxxx] All, Per our call last week, what are the thoughts on the wording of this as a proposed objective – Provide appropriate support for breakout functionality to 40G and / or 100G There was some concern about potential impact or unintended consequences that people wanted to see this wording to discuss further. Regards, John |