Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: Considerations for > 5 kms.

Anyone who feels qualified to put some thoughts together for these topics?


> -----Original Message-----
> From: Vipul Bhatt [mailto:vipul.bhatt@xxxxxxxxxxx]
> Sent: Tuesday, May 18, 1999 7:14 PM
> To: Jonathan Thatcher; stds-802-3-hssg@xxxxxxxx
> Subject: Considerations for > 5 kms.
> Jonathan,
> I would like to suggest a couple of topics for discussion, in case you
> haven't already included them in your agenda.
> 1. Use of WWDM in 1550 nm range for dark fiber, single mode, 
> metropolitan
> area applications: It seems to have some advantages. A 
> combination of DFB
> Isolated lasers and low attenuation will permit the link to 
> support useful
> distances (25 kms.?). If wavelengths are spaced widely enough, use of
> uncooled lasers will reduce cost and complexity of transmitters. These
> lasers can be directly modulated. If four wavelengths are 
> selected, each
> wavelength needs to carry data only at 2.5 Gbps (3.125 Gbaud 
> for 8B10B),
> which makes jitter issue more manageable compared to the 10 
> Gbps serial
> option.
> 2. Flow Control Implications: I am perplexed as to why no one 
> is talking
> about the impact of extended distances on the buffer size of 10GbE
> switches. Compared to Gigabit Ethernet, as we consider longer 
> distances and
> ten times the bit rate, it will take a much larger buffer in 
> a switch to
> operate properly. This is because a switch must accommodate 
> the greater
> number of bits that can still be in the pipeline after the 802.3x flow
> control is invoked. (Propagation delay through one kilometer 
> of fiber is
> the equivalent of more than 6 kilobytes of data at 10 Gbits/sec.) The
> longer the distance, the greater the buffer size required, or 
> the switch
> has to drop the packets. Until now, this hasn't been an issue because
> Ethernet was not expected to work at long distances so modest 
> buffer size
> was adequate. If we want 10GbE to work over MAN or even WAN 
> distances, the
> required increase in buffer size for switches may be so large 
> that at some
> point, packets have to be dropped. That may well be the basis of a
> cost-throughput tradeoff, but we should decide if we want to 
> address that
> tradeoff or leave it to implementers.
> Thanks,
> Vipul
> Vipul Bhatt
> Finisar Corporation
> 274 Ferguson Drive
> Mountain View CA 94043
> Phone: (650)691-4000 x113
> Fax: (650)691-4010
> Email: vipul.bhatt@xxxxxxxxxxx
> ========================