& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
<<   <   Page 12 / 12
Flower
Flower
12/5/2012 | 1:49:08 AM
re: Siemens Sees Ethernet Everywhere
What's your opinion on cut-through architectures vis-a-vis store-and-forward?

I'm not Tony, but I think I know the answer. :) Cut-through ethernet switched died when Kalpana was bought by cisco. After that, nobody bothered anymore.

Cut-through does not make sense on high-speed devices. When you have a 100 Mbps link, the time between transmission of the first and the last bit of a 4k frame is only a third of a millisecond. Hardly worthwile to try and improve only that 1/3 millisecond.

Besides that, cut-through switching only makes sense on lightly loaded networks. When link utilization goes up a bit, chances are that when a frame arrives on input interface X, another frame is being transmitted out on output interface Y. When that happens you need to store the frame in a buffer anyway, and fall back on store-and-forward.
sgan201
sgan201
12/5/2012 | 1:49:07 AM
re: Siemens Sees Ethernet Everywhere
Hi Seven,

DSL Modem ---UBR---> DSLAM ---> UBR ---> ATM switch---> VBR ---> ATM Switch--> BRAS

The VBR ATM VCC is not on the DSLAM..

Dreamer
sevenbrooks
sevenbrooks
12/5/2012 | 1:49:06 AM
re: Siemens Sees Ethernet Everywhere

ROFL, as I pointed out at the beginning of this. That model works in 1 and only 1 case: the DSLAM is unblocked. The moment it (or the DSL line) becomes a blocking element then the nice VBR VCC becomes USELESS. Traffic contracts must be maintained end to end. Its like feeding the Autobahn with 25 mile an hour roads and saying that the speed limit on the entire trip is unlimited.

So, when could this happen? Lets use Video over DSL as an example and use nice round numbers. Lets call every MPEG2 stream 3 Mb/s (this is actually low). So, 15 streams and a DS3 DSLAM is a blocking element (probably less than that). If an ADSL line is running at 8 Mb/s no more than 2 Video streams on a line. The line card may have separate blocking problems.

seven
Tony Li
Tony Li
12/5/2012 | 1:49:03 AM
re: Siemens Sees Ethernet Everywhere
Any pointers to appropriate buffer sizing? I guess too much doesn't help (hold on to a packet too long and there would be re-transmission), plus, it adds to latency. Most merchant silicon uses solely ingress buffering, leaving the device susceptible to HOL blocking.
---------------


Most folks tend to cite Van Jacobsen's work on high performance TCP that points to BW*RTT buffering on a per-link basis. There is some rational dissent on this issue, but my general feeling is that providing ample buffering and a knob to control its usage is the best of both worlds.


-------------

What's your opinion on cut-through architectures vis-a-vis store-and-forward?

-------------

Flower pretty much nailed it. Cut-through adds complexity to an architecture. In return, it decreases latency. Decreasing latency for some traffic, say local NFS, can have an effect. However, as soon as you get into non-trivial distances, you lose any benefit and Albert wins. Albert always wins. ;-)

Adding complexity without proportional benefit is a sophmoric mistake.

Tony
mr zippy
mr zippy
12/5/2012 | 1:49:00 AM
re: Siemens Sees Ethernet Everywhere
(Puts fire protectant suit on)

Adding complexity without proportional benefit is a sophmoric mistake.

IOW, ATM ?
PO
PO
12/5/2012 | 1:48:47 AM
re: Siemens Sees Ethernet Everywhere
Any pointers to appropriate buffer sizing? I guess too much doesn't help (hold on to a packet too long and there would be re-transmission), plus, it adds to latency. Most merchant silicon uses solely ingress buffering, leaving the device susceptible to HOL blocking.
---------------
Most folks tend to cite Van Jacobsen's work on high performance TCP that points to BW*RTT buffering on a per-link basis. There is some rational dissent on this issue, but my general feeling is that providing ample buffering and a knob to control its usage is the best of both worlds.


It's probably worth adding a comment that at least some of this rational dissent surrounds buffering closer to the network core (esp. long-haul links), where both BW and RTT are likely to be high. And it is less likely that there will be more than a single packet from any single flow in any single buffer simultaneously.

This, coupled with a reduced level of entropy (i.e. as more of the traffic flows together, it incurs the same buffers in sequence and experiences a 'leveling out' of the load) does suggest that there is room for more work on buffer sizing for that part of the network. It is likely, though, that Van's work can be extended macroscopically.
rpm
rpm
12/5/2012 | 1:48:45 AM
re: Siemens Sees Ethernet Everywhere
What's your opinion on cut-through architectures vis-a-vis store-and-forward?

Flower did pretty much nail the answer to this. 100 Mbps Ethernet killed most of the latency advantage that existed for 10/10 cut-through switching, plus when switching from a 10 Mbps port to a 100 Mbps port, it is necessary to store amd forward.

rpm
James_Donald
James_Donald
12/5/2012 | 1:47:48 AM
re: Siemens Sees Ethernet Everywhere
Please Ignore
<<   <   Page 12 / 12


Featured Video
Upcoming Live Events
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events
Partner Perspectives - content from our sponsors
Sports Venues: Where 5G Brings a Truly Immersive Experience
By Peter Linder, 5G Evangelist, North America, Ericsson
Multiband Microwave Provides High Capacity & High Reliability for 5G Transport
By Don Frey, Principal Analyst, Transport & Routing, Ovum
All Partner Perspectives