TokenStar: New LAN technolgoy (token-ring derived)

[SagaLore]

People warn me not to post ideas I have on the Internet out of fear of having someone else "steal" the idea - but hey, this is better than sitting on the idea and letting it die.

I've been conceptualizing a new network technology called Token Star for the last several years.

If you have ever researched Token Ring and dug into the flow control mechanisms rather than taking the physical wiring itself for granted, you'll agree with me that it was far superior over current Ethernet. It's downfall was IBM's unwillingness to consider that anything over 16mb was useful. I know that they were working on a fiber optic version of token ring but it was rediculously too pricey for most businesses.

Okay here it goes. I'll fix up my explanation later - this is a copy/paste from an IRC chat I had earlier.

<SagaLore> I have an invention I want to develop
<SagaLore> I'm going to bring back coax
<SagaLore> but it will be called TwinAx
<SagaLore> it is based on my TokenStar based LAN I'm also inventing
<SagaLore> it is basically the same as CoAx, but instead of a single straight copper wire, it is going to be a copper Twisted Pair
<SagaLore> I'm hoping for 100Gb speeds at long distances
<SagaLore> I'm stil pondering whether or not to incorporate a mesh shielding on it
<guy1> 100Gb speeds? 😵
<guy1> yeah if the network cable was directly hooked up to your ram
<guy1> or if you have a fricken server farm
<SagaLore> well by nature, TwinAx (and coax) is half duplex
<SagaLore> so its 100gb one way
<guy2> i get 2 TB speeds
<guy2> but that is only internal to my head
<guy2> the interface to the computer is much slower
<SagaLore> if you have a server supporting 100 machines, based on the TokenStar algorithm, only one machine at any time is transmitting
<SagaLore> the NICs would have a cache to queue up the transmissions that can take advantage of the fast speed and quickly send the Token Return packet
<SagaLore> so 100Gb comes in handy
<guy2> scalability?
<guy2> security?
<SagaLore> well
<SagaLore> TokenStar is going to fix most of the disadvantages to TokenRing
<SagaLore> Visually it won't be any different than an CAT5e Ethernet network
<SagaLore> Except that it will use a modification to the BNC connectors, the wire will be extremely thin
<SagaLore> So you can add cabling adhoc like you can with cat5e - from patch panel to wall port
<SagaLore> The TokenStar switch will use a different mechanism of switching - instead of store and forward - it will bebased on token control
<SagaLore> But the token will be controlled by the switch, not the NICs
<SagaLore> You would be able to easily create tokenstar VLANs this way
<SagaLore> You could also assign Token Priorities to ports - for example, every server in the data center needs to have the token 10 times for every workstation that gets a token
<SagaLore> Gives you more granular control over load balancing
<SagaLore> The idea I had was along with this network package, I would be pushing XTP/IPv6
<SagaLore> A few years ago I did report on 1Gb technologies and did a section on protocols written to support Gig and higher
<SagaLore> and XTP is a modification to TCP that is a lot more efficient at handling packets at those speeds
<SagaLore> it is also reverse compatible so you can run XTP on a network already supporting TCP

Discuss!

 

[scottws]

Seems like a noble idea, but very pie-in-the-sky.

 

[spidey07]

This sounds like token ring to me.

Token ring was a phyiscal star, but a logical ring toplogoy.

Also, what are you going to do for power? Devices expect power via their network connection these days.

But true, TCPs limitations on high speed networks need to be overcome. Also we already have 10, 40 G now and will be pushing for 100G soon.

-edit- almost forgot, there was at one time twinax cabling and probably still is in the video industry.

 

[SagaLore]

Originally posted by: spidey07
This sounds like token ring to me.

Token ring was a phyiscal star, but a logical ring toplogoy.

Also, what are you going to do for power? Devices expect power via their network connection these days.

But true, TCPs limitations on high speed networks need to be overcome.

Typically half the pairs in CATxx cabling get wasted. Some will try and use the other half to save on wiring costs (two wall ports), and then we came up with some devices like VoIP phones that can run low power over the other pairs. To me this seems like a waste and potential problems when trying to transmit data at high speeds.

Imagine the cost savings of wiring with this type of cabling. Once it competed mainstream with CAT5e, it would be a fraction of the cost. You could fit more in pipes, it would be more flexible, and less expensive to manufacture which is a savings that can trickle down to customers.

So I would say that for implementations where power is needed, you could run a partner cable along with the TwinAx - it would be straight copper running DC. I'm glad you brought this up - we could create the connector so it supports TwinAx along with the partner cable in a single connection.

TokenStar will have a logical star topology. The token handling happens at the switch rather than by the participants of the network. The reason I note that it is also a physical star is because the medium is based on coax which was a bus topology.

 

[spidey07]

well gigE and 10GigE over copper use all 4 pairs and provide power at the same time for cheap.

Also another problem that token ring had were the adapters - they were expensive because of the intelligence and token handling in hardware. scale that up to 100M, 1G, 10G and it may be cost prohibitive. Also, most token ring was run on cat 3/5 cable.

not trying to poke holes or shoot you down, just discussing.

And the fact that I'm glad that POS networking technology is dead. Beacon here, beacon there...beacon, beacon everywhere!
😀

 

[networkman]

I liked FDDI, but as spidey07 points out cost was a prohibitive factor once Ethernet took off.

 

[SagaLore]

Originally posted by: spidey07
-edit- almost forgot, there was at one time twinax cabling and probably still is in the video industry.

Oops. 😀 I guess I should have look up twinax before giving that the name of my cable.

Okay so what would be a good name for a cable with a single twisted pair? Twip? 😉

 

[cmetz]

SagaLore, how many actual physical network devices (Ethernet, Token Ring, or something other credible) have you designed and built?

 

[SagaLore]

Originally posted by: cmetz
SagaLore, how many actual physical network devices (Ethernet, Token Ring, or something other credible) have you designed and built?

I built a full duplex analog telecom system utilizing two tin cans and a solid plastic cable. 😉

Why do you ask? 😕

 

[Goosemaster]

Originally posted by: spidey07
well gigE and 10GigE over copper use all 4 pairs and provide power at the same time for cheap.

Also another problem that token ring had were the adapters - they were expensive because of the intelligence and token handling in hardware. scale that up to 100M, 1G, 10G and it may be cost prohibitive. Also, most token ring was run on cat 3/5 cable.

not trying to poke holes or shoot you down, just discussing.

And the fact that I'm glad that POS networking technology is dead. Beacon here, beacon there...beacon, beacon everywhere!
😀

OMFG spidey07...YOU UNPLUGGED YOUR PC:|

HURRY UP AND PLUG IT IN SO MY PC CAN CONTINUE DOWNLOADING PORN:|



:laugh:



LOFL..the good old days😎

 

[destrekor]

my only question is how would cat5e reach 100gbps?
and how will you overcome the passing of the token? on small networks the latency would be of no issue, and the token passing would be a great collision-prevention method instead of CSMA/CD used on ethernet, but is it feasible on large-scale networks?
and would current switches be able to support it? i only ask that because cisco, juniper, 3com, etc etc likely do not want to invest into anything other than ethernet for LAN technologies anymore. everyone feels ethernet is the way to go, to bringing in something else likely would be shot down.
could/did any form of tokenring operate at ethernet's layer3 protocols, or would/did they have their own layer3 protocols? if it could be worked into using the same protocols (if even possible) then it could have a successful future. but as I said, it could be hard getting the technology ratified.

 

[Zstream]

Originally posted by: SagaLore
People warn me not to post ideas I have on the Internet out of fear of having someone else "steal" the idea - but hey, this is better than sitting on the idea and letting it die.

I've been conceptualizing a new network technology called Token Star for the last several years.

If you have ever researched Token Ring and dug into the flow control mechanisms rather than taking the physical wiring itself for granted, you'll agree with me that it was far superior over current Ethernet. It's downfall was IBM's unwillingness to consider that anything over 16mb was useful. I know that they were working on a fiber optic version of token ring but it was rediculously too pricey for most businesses.

Okay here it goes. I'll fix up my explanation later - this is a copy/paste from an IRC chat I had earlier.

<SagaLore> I have an invention I want to develop
<SagaLore> I'm going to bring back coax
<SagaLore> but it will be called TwinAx
<SagaLore> it is based on my TokenStar based LAN I'm also inventing
<SagaLore> it is basically the same as CoAx, but instead of a single straight copper wire, it is going to be a copper Twisted Pair
<SagaLore> I'm hoping for 100Gb speeds at long distances
<SagaLore> I'm stil pondering whether or not to incorporate a mesh shielding on it
<guy1> 100Gb speeds? 😵
<guy1> yeah if the network cable was directly hooked up to your ram
<guy1> or if you have a fricken server farm
<SagaLore> well by nature, TwinAx (and coax) is half duplex
<SagaLore> so its 100gb one way
<guy2> i get 2 TB speeds
<guy2> but that is only internal to my head
<guy2> the interface to the computer is much slower
<SagaLore> if you have a server supporting 100 machines, based on the TokenStar algorithm, only one machine at any time is transmitting
<SagaLore> the NICs would have a cache to queue up the transmissions that can take advantage of the fast speed and quickly send the Token Return packet
<SagaLore> so 100Gb comes in handy
<guy2> scalability?
<guy2> security?
<SagaLore> well
<SagaLore> TokenStar is going to fix most of the disadvantages to TokenRing
<SagaLore> Visually it won't be any different than an CAT5e Ethernet network
<SagaLore> Except that it will use a modification to the BNC connectors, the wire will be extremely thin
<SagaLore> So you can add cabling adhoc like you can with cat5e - from patch panel to wall port
<SagaLore> The TokenStar switch will use a different mechanism of switching - instead of store and forward - it will bebased on token control
<SagaLore> But the token will be controlled by the switch, not the NICs
<SagaLore> You would be able to easily create tokenstar VLANs this way
<SagaLore> You could also assign Token Priorities to ports - for example, every server in the data center needs to have the token 10 times for every workstation that gets a token
<SagaLore> Gives you more granular control over load balancing
<SagaLore> The idea I had was along with this network package, I would be pushing XTP/IPv6
<SagaLore> A few years ago I did report on 1Gb technologies and did a section on protocols written to support Gig and higher
<SagaLore> and XTP is a modification to TCP that is a lot more efficient at handling packets at those speeds
<SagaLore> it is also reverse compatible so you can run XTP on a network already supporting TCP

Discuss!

Cat5e, cat6, cat 7 cannot support 100gb transmissions no matter what topology you used.

 

[SagaLore]

Originally posted by: Zstream
Cat5e, cat6, cat 7 cannot support 100gb transmissions no matter what topology you used.

Correct, which is why I'm inventing a new kind of cable...

Originally posted by: destrekor
my only question is how would cat5e reach 100gbps?
and how will you overcome the passing of the token?

Token control would occur at the Switch, not on each node (NIC card). When a node (workstation or server) wants to transmit data, it sends a Token Request to the Switch. The Switch then adds the port to the Star. When the node is done, it sends a Token Return back to the Switch, which then drops the port from the Star. Because the Switch negotiates all communications between nodes, no single node will ever disrupt the Star.

 

[spidey07]

Originally posted by: destrekor
my only question is how would cat5e reach 100gbps?
and how will you overcome the passing of the token? on small networks the latency would be of no issue, and the token passing would be a great collision-prevention method instead of CSMA/CD used on ethernet, but is it feasible on large-scale networks?
and would current switches be able to support it? i only ask that because cisco, juniper, 3com, etc etc likely do not want to invest into anything other than ethernet for LAN technologies anymore. everyone feels ethernet is the way to go, to bringing in something else likely would be shot down.
could/did any form of tokenring operate at ethernet's layer3 protocols, or would/did they have their own layer3 protocols? if it could be worked into using the same protocols (if even possible) then it could have a successful future. but as I said, it could be hard getting the technology ratified.

CSMA/CD is no longer a concern for ethernet. none whatsoever.

Zstream, I believe you will see 100 gig ethernet on category rated cable. We though it would be impossible to do 10G, but they do.

 

[qaa541]

Originally posted by: SagaLore

Originally posted by: cmetz
SagaLore, how many actual physical network devices (Ethernet, Token Ring, or something other credible) have you designed and built?

I built a full duplex analog telecom system utilizing two tin cans and a solid plastic cable. 😉

Why do you ask? 😕

That phone system is not full duplex! Hello collisions! 😉

 

[spidey07]

Originally posted by: SagaLore

Originally posted by: Zstream
Cat5e, cat6, cat 7 cannot support 100gb transmissions no matter what topology you used.

Correct, which is why I'm inventing a new kind of cable...

Originally posted by: destrekor
my only question is how would cat5e reach 100gbps?
and how will you overcome the passing of the token?

Token control would occur at the Switch, not on each node (NIC card). When a node (workstation or server) wants to transmit data, it sends a Token Request to the Switch. The Switch then adds the port to the Star. When the node is done, it sends a Token Return back to the Switch, which then drops the port from the Star. Because the Switch negotiates all communications between nodes, no single node will ever disrupt the Star.

But that's the biggest problem. All that request/response at layer2 would add latency plus signficant hardware constraints operating at those speeds. Modern networks just let end nodes scream all they want and buffering/queuing/QoS takes care of the rest. Modern networks have 1 Megabyte per port of buffering capacity with 6-8 hardware queues - per port.

Its one of the reasons why ATM failed. It sounded like such a great technology, but it is outdated.

 

[qaa541]

How would twisted pair coax work? The coax cables are already shielded independently, so the twisted pair effect wouldnt work if I'm interpreting your idea right. I understand your idea as two independent coax cables twisted together. If you mean two copper conductors twisted together and then sheilded with a metal conductor, then I dont know the outcome of that scenario (I'm no RF expert), but I still don't think it will give the kind of speeds you are looking for.

My bets are on multi-wavelength optical for high speed multigig networks. There is alot less interference with optical cabling.

 

[SagaLore]

Originally posted by: spidey07
Zstream, I believe you will see 100 gig ethernet on category rated cable. We though it would be impossible to do 10G, but they do.

Currently 10G has to run over a modified CAT6 which has every pair shielded. And it is expensive.

 

[spidey07]

Originally posted by: SagaLore

Originally posted by: spidey07
Zstream, I believe you will see 100 gig ethernet on category rated cable. We though it would be impossible to do 10G, but they do.

Currently 10G has to run over a modified CAT6 which has every pair shielded. And it is expensive.

you are incorrect. That would be cat7.

-edit- what I should say is all the cable manufacturers are chomping at the bit to sell you new cable while the IEEE is wanting to make it work over existing cabling.

It's been a few months, but I have had every major cable company pitch me their spiel on how they are going to do 10G and the pros/cons. It's all marketing fluff because AFAIK the standard isn't done yet.

 

[spidey07]

Originally posted by: qaa541
How would twisted pair coax work? The coax cables are already shielded independently, so the twisted pair effect wouldnt work if I'm interpreting your idea right. I understand your idea as two independent coax cables twisted together. If you mean two copper conductors twisted together and then sheilded with a metal conductor, then I dont know the outcome of that scenario (I'm no RF expert), but I still don't think it will give the kind of speeds you are looking for.

My bets are on multi-wavelength optical for high speed multigig networks. There is alot less interference with optical cabling.

But the current push is for copper. Fiber already has more than it needs for a cable, its the optics driving it that are the limitation.

 

[SagaLore]

Originally posted by: spidey07
But that's the biggest problem. All that request/response at layer2 would add latency plus signficant hardware constraints operating at those speeds. Modern networks just let end nodes scream all they want and buffering/queuing/QoS takes care of the rest.

How would that be any less efficient than a 3-way handshake?

Originally posted by: qaa541
How would twisted pair coax work? ... If you mean two copper conductors twisted together and then sheilded with a metal conductor, then I dont know the outcome of that scenario (I'm no RF expert), but I still don't think it will give the kind of speeds you are looking for.

The idea behind twisting the pairs is awesome. It is a way of canceling out interference. UTP won over STP not just because it was cheaper to make, but because they found it was more reliable. It takes a very simple circuit to clear up the signal. But when you add shielding and throw current down the middle of it, you basically create an antennae which actually makes the signal worse.

 

[spidey07]

Originally posted by: SagaLore

Originally posted by: spidey07
But that's the biggest problem. All that request/response at layer2 would add latency plus signficant hardware constraints operating at those speeds. Modern networks just let end nodes scream all they want and buffering/queuing/QoS takes care of the rest.

How would that be any less efficient than a 3-way handshake?

different layers, different things that they are trying to do/responsible for.

Layer 2 is "get bits framed properly to destination"
Layer 3 is "get packet to destination"
Layer 4 is "let's setup a reliable connection independant of those bastages layer2 and layer3"

 

[spidey07]

PM to scottmac sent.

 

[SagaLore]

Originally posted by: spidey07
you are incorrect. That would be cat7.

Yea sorry you're right - I did a quick search before I posted and there are a few vendors that make a shielded cat6 for 10G but like you said, it isn't standard.

Originally posted by: spidey07

Originally posted by: SagaLore
How would that be any less efficient than a 3-way handshake?

different layers, different things that they are trying to do/responsible for.

Layer 2 is "get bits framed properly to destination"
Layer 3 is "get packet to destination"
Layer 4 is "let's setup a reliable connection independant of those bastages layer2 and layer3"

TCP over Ethernet:

Syn, Syn/Ack, Ack, open session, Fin

TokenStar:

Token Request, Tocken Granted, open session, Token Return

It's almost exactly the same, except there is much more order going on in the Switch.

 

[spidey07]

Originally posted by: SagaLore

Originally posted by: spidey07
you are incorrect. That would be cat7.

Yea sorry you're right - I did a quick search before I posted and there are a few vendors that make a shielded cat6 for 10G but like you said, it isn't standard.

be wary of cable guys pushing 10G over copper. right now its all fluff.

I got to grill one of Siemens design engineers and needless to say she was speechless.

but if you follow the progression of ethernet, it will come. 10G cheaply over copper will come.