non-Spanning Tree ways of tolerating loops

Are there ways other than Spanning Tree Protocol to handle loops in a
switching network. For reasons of low latency I want to minimize the number
of switch hops required peer-to-peer on my topology. The best way seems to
be a mesh or looped network topology.

Unfortunately, STP just logically removes physical loops. Are there ways to
actually use loops to get shortest paths but do away with the negatives
like broadcast storms?

--
Rahul

Rahul <(E-Mail Removed)> wrote:

> Are there ways other than Spanning Tree Protocol to handle loops in
> a switching network. For reasons of low latency I want to minimize
> the number of switch hops required peer-to-peer on my topology. The
> best way seems to be a mesh or looped network topology.

What order of magnitude of latency are you dealing with presently?
How many peers are you dealing with?

> Unfortunately, STP just logically removes physical loops. Are there
> ways to actually use loops to get shortest paths but do away with
> the negatives like broadcast storms?

Token Ring and source routing?-)

--
portable adj, code that compiles under more than one compiler
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

Rick Jones <(E-Mail Removed)> wrote in news:hm1h41$rhh$1
@usenet01.boi.hp.com:

>> Are there ways other than Spanning Tree Protocol to handle loops in
>> a switching network. For reasons of low latency I want to minimize
>> the number of switch hops required peer-to-peer on my topology. The
>> best way seems to be a mesh or looped network topology.
>
> What order of magnitude of latency are you dealing with presently?

Around 13 microsecs end-to-end. But each extra switch hop adds ca. 3
microsecs. So reduction in hops could be significant.

> How many peers are you dealing with?

~300 servers. Each switch can take ca. 40. So ~8 switches or more depending
on topology. Currently worst-case 3 switches might be there in the path.
But if I allow loops only 2 switches will be worst case.


>> Unfortunately, STP just logically removes physical loops. Are there
>> ways to actually use loops to get shortest paths but do away with
>> the negatives like broadcast storms?
>
> Token Ring and source routing?-)

So nothing in the ethernet domain? I have never used a token ring
framework, are those still around? Does that need special Hardware or will
my Linux+ethernet card + commodity switch framework handle Token Ring?


--
Rahul

Rahul <(E-Mail Removed)> wrote:
> Rick Jones <(E-Mail Removed)> wrote in news:hm1h41$rhh$1
> @usenet01.boi.hp.com:

> >> Are there ways other than Spanning Tree Protocol to handle loops in
> >> a switching network. For reasons of low latency I want to minimize
> >> the number of switch hops required peer-to-peer on my topology. The
> >> best way seems to be a mesh or looped network topology.
> >
> > What order of magnitude of latency are you dealing with presently?

> Around 13 microsecs end-to-end. But each extra switch hop adds ca. 3
> microsecs. So reduction in hops could be significant.

It is only changing a constant, but the quickspecs for the HP ProCurve
6120 claim 1.8 microseconds latency, and my netperf TCP_RR tests
comparing back-to-back with through that switch suggest 1.9. The 6120
is not all that big/high-end a switch so perhaps there are others with
similar latency?

> > How many peers are you dealing with?

> ~300 servers. Each switch can take ca. 40. So ~8 switches or more
> depending on topology. Currently worst-case 3 switches might be
> there in the path. But if I allow loops only 2 switches will be
> worst case.


> >> Unfortunately, STP just logically removes physical loops. Are
> >> there ways to actually use loops to get shortest paths but do
> >> away with the negatives like broadcast storms?

> > Token Ring and source routing?-)

> So nothing in the ethernet domain? I have never used a token ring
> framework, are those still around? Does that need special Hardware
> or will my Linux+ethernet card + commodity switch framework handle
> Token Ring?

It was meant as a joke I think that even IBM stopped Token Ring
work at 100Mbit/s speeds.

You might ask in comp.dcom.lans.ethernet

rick jones
--
the road to hell is paved with business decisions...
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

Rick Jones <(E-Mail Removed)> wrote in news:hm1vgo$uhi$4
@usenet01.boi.hp.com:

> It was meant as a joke I think that even IBM stopped Token Ring
> work at 100Mbit/s speeds.

Good! I thought I had to relearn a technology that I thought was long
dead. Nothing like a little ignorance to distroy a good joke.

--
Rahul

David Schwartz <(E-Mail Removed)> wrote in news:2fbee1c9-343e-4fe7-
981d-(E-Mail Removed):

> There are tons of ways. They all have trade-offs, of course.
>

Thanks David. But are any of these ways included in standard commodity
hardware? Have they made it to protocols? Or are these more of academic
curiosities? It would be great if something like this was implimentable
under ethernet.

I got a sugesstion to look up OSPF. Apparantly many common switches seem to
support that. Not sure if that does exactly what I need though. They seem
to do "wire speed IP routing" but I thought this was a "switching problem".
Not sure how routing enters the picture.

--
Rahul

David Schwartz <(E-Mail Removed)> wrote in news:cd2d139f-f4fb-4ca0-
bbde-(E-Mail Removed):

> Since this can't be done at the switching level, if you really want
> it, do it at the routing level. Algorithms like OSPF allow all the
> links to be active, support multiple routes to the same destination,
> and pick a sensible path for each packet.

Can rounting be done as fast as switching though? For cases where latencies
are critical? Is there a cut-through mode for routing as well?

Also most commodity equiment I've worked with is sold as a "switch". Is
there anything different about buying a "real" router or are routing and
switching pretty similar operations for the hardware buying part?

As I understood it switches usually used some kind of ASIC whereas routers
were microprocessor based (my elementary wikipedia knowledge)



> IMO, large scale switching is a bad idea anyway. You don't want 1,500
> hosts in the same Ethernet broadcast domain.
>

I thought routers were used when 2 or more subnets were involved. Here I
have 300 hosts but all on the same subnet. Can I still use a router or do
you mean 300 hosts on one subnet is a bad idea too?

--
Rahul

Using routers means having multiple IP subnets.

Take your 300 hosts, divide them into 10 subnets of 30 hosts each.
Get you hands on 11 switches.

Logically, tour building blocks are 30 nodes, a "router" and a switch
(well, two switches) Connect the 30 nodes and "router" to the switch,
lather, rinse, repeat 9 times. Now connect the 10 "routers" to the
11th switch.

Of course, at a block diagram level, that is still three hops - two
switch hops and a "router" hop.

rick jones
--
I don't interest myself in "why." I think more often in terms of
"when," sometimes "where;" always "how much." - Joubert
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

Rick Jones <(E-Mail Removed)> wrote in news:hm43tl$dkn$1
@usenet01.boi.hp.com:

> Using routers means having multiple IP subnets.
>
> Take your 300 hosts, divide them into 10 subnets of 30 hosts each.
> Get you hands on 11 switches.
>
> Logically, tour building blocks are 30 nodes, a "router" and a switch
> (well, two switches) Connect the 30 nodes and "router" to the switch,
> lather, rinse, repeat 9 times. Now connect the 10 "routers" to the
> 11th switch.
>
> Of course, at a block diagram level, that is still three hops - two
> switch hops and a "router" hop.

That's a lose-lose situation for me then. I wanted loops because a looped
network (mesh really) minimizes my hops. [think of as many switches as
needed to connect all servers symetrically but where every switch is in
addition connected directly to every other switch ]

But if I put in routers and increase my hops I might as well stay with the
original switched network anyways with all servers on a single switched
subnet with no loops. If I have no loops I need no routers. I wanted loops
to minimise hops.

Its circular.

--
Rahul

David Schwartz <(E-Mail Removed)> wrote in
news:21b348f1-9caa-4acd-9dac-(E-Mail Removed):

> es
>> are critical? Is there a cut-through mode for routing as well?
>
> Cut-through switching is obsolete.

Why? What's wrong with cut-through? I don't need deep packet inspection
abilities. Cut-through seems to work well.

> Exactly how critical are latencies?

Very critical. More critical than anything else. More critical than
bandwidth. More critical than reliability or redundancy. Don't need any
advanced switch magic. No QoS, no security, no link aggregation etc. Just
fast packet switching.

> What size network are we talking about?

300 servers. About 40 ports per switch.

> They are essentially the same. Just make sure you get a routing switch
> that can route at wire speed.

If a switch can switch at wire speed does it imply that it can route at
wire speed too?

> There are many that can, but also many
> that can't. You have an unusual application, so will have to look for
> reviews and network tests (ideally those that measure latency) and you
> won't be able to buy whatever switch is cheapest.

I have already tested switches. I know what latencies each offers for
switching. But I still need to minimize hops by optimal network design.

Also, my latency tests were in switching mode. Will they stay the same in
router mode? Isn't routing additional work for the ASICs?

> Switches and routers both now use an ASIC for the actual forwarding.
>
>> > IMO, large scale switching is a bad idea anyway. You don't want
>> > 1,500 hosts in the same Ethernet broadcast domain.

What if my traffic is such that broadcasts are not a huge fraction of
traffic?

>> Here I
>> have 300 hosts but all on the same subnet.
>
> Why?

Because it allows me to have a switched only network. I thought switching
was faster than routing. And to us latencies are critical.

>> Can I still use a router or do
>> you mean 300 hosts on one subnet is a bad idea too?
>
> It's not a bad idea if you can manage them without needing features
> that only routing can provide.

To me it's just one feature that routers will be better at: ability-to-
tolerate loops.

>But it sounds like you need the ability
> to route between them, so that would make it a bad idea.

It's all about performance. If the router degrades latencies by larger
amounts than the loops would have improved I have to just stay switched
but loop-less.

--
Rahul