CRC Errors on GigE.

A possible explanation would be that Ciscos PHY might produce a better Signal/Noise ratio for outgoing symbols than Dells. So frames coming in with FCS errors originating from one switch would be error-free when sent from another.

Reply to
Denis Jedig
Loading thread data ...

For short cables there should be plenty of signal from any device. I might believe S/N for longer cables.

-- glen

Reply to
glen herrmannsfeldt

Even under specific environmental conditions like borderline electromagnetic disturbances (induced by fluorescent tubes or electromechanic equipment nearby the cabling)?

Reply to
Denis Jedig

Mostly, yes. Someone did ask once about arc welder, and I will make an exception for them. Attenuation is exponential, and much worse at higher frequencies.

-- glen

Reply to
glen herrmannsfeldt

;-) Does anybody have any characteristic graphs of typical signal distortions caused by electric and electromechanic equipment among the frequencies used by ethernet in UTP/STP cable? Arc welders need not necessarily be present on this list.

BTW, another idea for the possible cause of FCS errors is the far-end crosstalk or even reflection. Since the cable length is short, the attentuation for FEXT and RL is low as well, so maybe the PHY, the interconnections or the RJ45 jack of the Dell switch are somewhat imbalanced[1] inducing FCS errors on frames leaving the PHY in this very moment.

[1] granted, they will have to be pretty much imbalanced to shift a symbol
Reply to
Denis Jedig

Here are a few ideas and suggestions that I think are pertinent to the discussion.

0=2E Modern network links appear to be astonishigly reliable.

Here is the first Gigabit copper link with big numbers that I looked at.

It is between a Cisco 4500 and a 6500 a few feet apart.

uptime is 22 weeks, Input queue: 0/2000/0/0 (size/max/drops/flushes); Total output drops:

952 156,978,733,398 packets input, 147168474300246 bytes, 0 no buffer Received 2584729806 broadcasts (1702867892 multicasts) 0 runts, 0 giants, 0 throttles 0 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored

Note that network congestion is MUCH more prevelant than hardware errors.

Zero errors per year are not at all unusual even on quite busy links on copper and is expected on fiber.

1=2E Dell are not a switch supplier I would consider for anything. I know nothing of their switches and plan to keep it that way for professional use. I have nothing against Dell as such, it's just that the network is often so important that using stuff you know and trust and can get support on can be worth a lot.

2=2E Have you considered doing your own Error Correction? e=2Eg. send two copies of every packet. Send them over seperate networks?

There are I am sure many options. Send more smaller packets and do some sort of error correction that can stand the loss of a whole (or more) packet(s). Send less data to compensate. How much /Information/ is being sent is each packet? e.g Are you sending

32bit integers representing boolean values?

Consider investigating RAID technology, Reed-Solomon encoding. Hamming code.

3=2E /Almost/ all or all modern network switches are store and forward. This /will/ be costing you an extra serialisation delay for each hop. I have some Cisco experience and as far as I know there are /no/ network switches made by Cisco that can be used in a cut-through mode today. You say 500 bytes in 4us but in fact this will be being transmitted twice, one by the source and once by the switch resulting in an effective serialisation delay of 8us plus any delay caused by the forwarding process. The latter however is likely to be approximately zero.

4=2E I suspect that Cut-through Gigabit switches are available. A quick google turns up some chipsets. (Make your own:-) You'll need one of these:-

formatting link

5=2E There exist standards for networks that are used by the military. When the bomb rack gets a message to let go it needs to be pretty sure that nothing is broken so maybe reliability is built in? Probably not so fast though.

6=2E Andrew Tanenbaum (Computer Networks) Says:- [Very roughly] There is quite a decent chance that a bad frame will have a good CRC. However your frame rate is so low that it doesn't seem as if that will matter. Mathematics are in the book.

I read the Second Edition a decade or so ago.

formatting link
Networks, Fourth Edition (=A92003) ISBN: 0-13-066102-3

7=2E For long term reliable operation you need to know how your LAN switches work. e=2Eg. Will normal or abnormal operation ever put the switch into Fast Aging? Do frames get delayed when the destination MAC is unknown.

Are you keeping the forwarding database populated?

I would be astonished of Rich Seifert's "Switch Book" book was not the best place to go for this and may cover the CRC in detail too.

The Switch Book: The Complete Guide to LAN Switching Technology Rich Seifert ISBN: 978-0-471-34586-2

I would be tempted to avoid the lowest end of the market even for a supplier like Cisco.

8=2E There may be switches available that use ECC memory and ECC busses/fabrics. No idea.

9=2E Sounds to me like you need a network consultant on board. Ask Rich Seifert, but maybe he is too busy being a Lawyer now.

  1. There is always 10GE.

Hope this may assist.

Reply to
Bod43

While I *have* recently been admitted to the California bar, I am not

*actively* practicing law. I am, however, spending an inordinate amount of time on the tennis court. ;^)

I still consult on technology issues, but rarely get involved with end-user network design or troubleshooting, unless the situation is

*really* interesting or fun. For example, I helped design the network used for online transaction processing on the trading floor of the NY Stock Exchange (availability is paramount, money is no problem, and any mistakes show up on the front page of the Wall Street Journal), and more recently helped debug a performance problem on a LAN located on the International Space Station (long lead-time on spare parts, diagnostic and repair instructions must be written for astronauts, and all work must be cleared through mission control). My kind of fun.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

Reply to
Rich Seifert

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.