Yes, after reading a couple of articles I see that WDM is what I want to do. Do you need separate hardware for this or can you plug a GBIG in a switch or a network module in a router router to get this capability? (I have a 2924 and 3640.)
-- Bob Simon remove x from domain for private replies
Thanks for the pointer to this product. This would work for my needs but I'd have to replace my 2924 with a 2950 or 3550. I'm going to try to price out the options to see what a total solution would cost.
If anyone else can suggest other alternatives, I'd certainly like a few other options to present to my boss.
-- Bob Simon remove both "x"s from domain for private replies
You'd certainly need separate hardware on both ends for handling DWDM (or WDM to that matter) that would have prisms to break the light down to separate wavelengths and then combine it back. These are bulky and will not fit into your regular GBIC footprint, at least for now. They will definitely make them smaller in the future, but not yet as a commercial product.
If you buy a simple Ethernet multiport bridge, normally now referred to as "layer 2 switch," the Ethernet frames from any of the ports will be sent to any other port as required. Sometimes they are flooded to all ports (e.g. in the case of Ethernet broadcasts such as ARP queries), other times they are sent to the port that reaches the destination host.
So by definition, a switch should do what you ask.
However, if the Ethernets you are "multiplexing" each belong to a unique IP subnet, then the box you need is not a layer 2 switch but rather a router, also known as "layer 3 switch." One of the router ports can be connected to the long link.
The 1400 ft length should be okay for multimode 62.5/125 um fiber even with Gigabit Ethernet, but only at the 1300 nm wavelength. If you use
850 nm, you're limited to about 720 ft. At lower speeds, e.g. 10 or 100 Mb/s, no problem meeting that length requirement with short or long wavelength.
If you expect to migrate to really fast stuff, like 10Gb/s Ethernet, use of multimode fiber becomes more complicated. It is possible with some combinations out to 300 meters (984 ft.). Probably safer to stick with single-mode fiber in the long links, if you expect to go to 10G.
You haven't said what total bandwidth you require. If it is < 1G bps in each direction then it is easy and cheap. You just use 1000BaseLX (GBICS, SMF) and VLAN trunking. You will need a mode conditioning cable for each end too.
"This would work for my needs but I'd have to replace my 2924 with a 2950 or 3550."
If it is less than 100Mbps then you can use your existing switches with media converters which would be even cheaper.
Page > Ethernet Converters > D-Link DFE-857 Fast Ethernet Media Converter Our Price: £383.45 to choose the first one I came across. That seems a bit expensive:-
I would check that ther were compatible with your choice of trunking protocol 802.1q or (less likely now ISL) since that increases the frame size. Surprisingly enough they seem to be fairly reliable.
If you need more bandwidth than 1Gbps then you are into WDM territory of which I have no experience but I would be surprised if anyone did Multi Mode WDM kit.
Note that LX GBICS are single Mode and can be used with 62.5/125 cable. Maybe other single mode technologies (WDM) would work too? However I imagine that there would be no support available even if it did work.
Several folks have recommended VLANs. I would really prefer greater isolation between streams but VLANs may work for me until I run out of bandwidth. With gigabit, I hope we'll be able to defer CWDM for at least a year. Although I don't have SMF, Albert mentioned that I'll be able to run gigabit 1400' over MMF at 1300 nm.
Are VLANs protocol independent? What is a mode conditioning cable?
I'll research the various types and capabilities of GBICs this afternoon.
Thanks but I would not be comfortable deploying a solution designed for SMF over MMF. Network stability is critical.
-- Bob Simon remove both "x"s from domain for private replies
If ten 100 Mb/s Ethernets share a single 1 Gb/s trunk segment, there will be no possibility of congestion in the trunk. Even if 20 such Fast Ethernets share the same trunk, congestion will most likely be completely manageable. It all depends on what type of traffic uses this infrastructure. For data traffic, no sweat.
In reading the original post, I got no indication that there was anything unusual about providing such a trunk cable for Ethernet segments. Any kind of wave division multipexing is normally only for really fancy WAN infrastructure. I was surprised to see WDM even come up in this example.
And the point is too that packet switched systems can naturally perform "statistical multiplexing." In networks that have to support primarily bursty data traffic, it would be very rare that all of the Ethernets sharing that one trunk cable would experience peaks at exactly the same time. This even holds true if the Ethernets are supporting compressed video or audio streams.
So I wouldn't automatically assume that the trunk segment *must* have a bandwidth equal to the simple sum of all Ethernet segments feeding it. In most cases, that would be a very wasteful design.
so long as the traffic is well behaved - but if each of those Ethernets already uses VLANs, then you need label stacking (802.1q in q i think cisco call it)
i have had hassles with the same MAC address in different VLANs on some low end switches before upsetting the shared forwarding tables as well.
Given that those issues are unusual and / or avoidable you should be OK.
i dont think WDM will be acceptable cost for now (although we dont know what are the issues, but MMF sounds like local cabling in a campus).
i have used CWDM GBICs in Nortel and Cisco gear - it works pretty well (this was on rented single mode fibre, so we were trying to cut the number of cores needed).
it gives you up to 8 lambdas, and works with passive glass mux / demuxes (4 or 8 interfaces) - you can also do drop and insert, but you loose some signal through each mux.
the drawback is that this stuff has reasonably high power lasers, since the whole idea is to cut the cost of adding extra fibres - Cisco list their GBICs at around $5k each....
but there are much cheaper sources such as Finisar (and neither Cisco nor Nortel make their own, but some cisco gear checks for "accepatble" GBIC type IDs).
1 thing no one has mentioned is using an Ethernet "mux" - some low end SDH muxes such those from Axxessit in Norway may be a better bet, but we are still talking $1000s for equipment - so it isnt going to compete with a bit more fibre in the duct, or Gigabit and VLANs.