Sign in with
Sign up | Sign in
Your question

TX, fully switched -- hop limit myths?

Last response: in Networking
Share
Anonymous
May 20, 2004 9:44:52 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

I have been faithfully evolving our LAN within the hop-count limits
that I have read about in newsgroups and online -- namely that
one is permitted 4 intermediate hops between end-stations for 10 Mbps,
and 3 intermediate hops between end-stations for 100 Mbps, with
each router "resetting" the count.

Several months ago, I went to check the exact name of the "rule"
(which I knew as the "3-4-5 rule"), and discovered that the "5"
referred to permitted fan-outs in a star topology; and I knew from
my readings and from practical experience that fan outs beyond 5
are extremely common (e.g., 24 or 48 port switches, potentially
stacked into the hundreds of ports range.)

At that point I started to doubt that the rule was for the same -kind-
of LAN that is commonly deployed these days, and the more I thought about
those "rules", the more I have come to doubt that what I had always
"known" to be a hard "Thall Shalt Not" rule had any meaning for our
network.

I did some digging in the 802.3 specs last night, and I don't find
that "rule" anywhere, and what I did find seems to lend credence
to my thought processes. I would appreciate some confirmation or
refutation from those who know ethernet better than I.


The limits I found in 802.3 were on Class I and Class I repeaters
(which I tend to think of as 'hubs'). I found a sentance in the
100BaseTX section that indicated that when you are using the
classic star topology with duplex links with bridges between the
segments, then each segment is treated as an independant collision
domain, and the segment size limit is then bound just by the 100 metres
limit rather than by round-trip propogation limits. But then a later
sentance talks about limits on the "network" without any proximate
wording to clarify if "network" is referring to a single
collision domain.


Am, then, I now properly interpreting that in the case where every device
is connected to a switch port through a proper Cat5 cable of < 100 metres
each, and the switches are themselves connected together through
proper Cat5 < 100 metres (or through 100BaseFX or 1000BaseFX if over 100m),
then it is permitted to have an arbitrary number of switches in the
chain [with the higher level protocol timers then becoming the
operative limits]?


If that -is- true, that there -is- no 3 hop rule at 100BaseTX /
1000BaseTX that is meaningful for fully switched network, then
certain topological changes I have simmering would be much simplified.
I'm planning to migrate some equipment to a new gigabit switch that
doesn't stack with our existing multi-vlan'd 100 Mb swithces; if there
is no hop count limit, then the new device becomes simple to introduce,
and transition to; if there -is- a maximum of 3 switches before hitting
a router, then I'll have to do noticable internal renumbering to
ensure that no -actual- path exceeds 3 switches before routing.
--
The Knights Of The Lambda Calculus aren't dead --this is their normal form!
Anonymous
May 20, 2004 9:44:53 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

Hello, Walter!
You wrote on 20 May 2004 17:44:52 GMT:

WR> If that -is- true, that there -is- no 3 hop rule at 100BaseTX
WR> /
WR> 1000BaseTX that is meaningful for fully switched network, then
WR> certain topological changes I have simmering would be much
WR> simplified.

Default STP timers (802.1d) are based on assumption that the network diameter is
7 hops max. That doesn't mean though that big flat L2 networks is a good thing.
There is three very useful chapters in Cisco Lan Switching by Clark and
Hamilton - 14. Campus Design Models, 15. Campus Design Implementation and 17.
Case Studies.

WR> I'm planning to migrate some equipment to a new gigabit switch
WR> that doesn't stack with our existing multi-vlan'd 100 Mb
WR> swithces; if there is no hop count limit, then the new device
WR> becomes simple to introduce, and transition to; if there -is-
WR> a maximum of 3 switches before hitting a router, then I'll
WR> have to do noticable internal renumbering to ensure that no
WR> -actual- path exceeds 3 switches before routing.

Is this new gigabit switch has a layer 3 functionality? Modern Cisco switches -
3550, 3750, 4500 (with some line cards), 6500 (Sup 720 with some line cards) -
has non-blocking backplane and can do wirespeed routing. It would be worthwile
to check if you can use it as a core/distribution switch.

With best regards,
Andrey.
Anonymous
May 20, 2004 10:16:56 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In comp.dcom.lans.ethernet Walter Roberson <roberson@ibd.nrc-cnrc.gc.ca> wrote:
> one is permitted 4 intermediate hops between end-stations for 10 Mbps,

I know this one, 500m collision diameter with some (generous?)
allowance for equipment propagation delay.

> and 3 intermediate hops between end-stations for 100 Mbps, with

Never heard of this. AFAIK when using 100baseTX hubs, you
get a short 5m interconnect two (depends on hub mfr) for a
maximum collision diameter of 205m.

> Am, then, I now properly interpreting that in the case where every device
> is connected to a switch port through a proper Cat5 cable of < 100 metres
> each, and the switches are themselves connected together through
> proper Cat5 < 100 metres (or through 100BaseFX or 1000BaseFX if over 100m),
> then it is permitted to have an arbitrary number of switches in the
> chain [with the higher level protocol timers then becoming the
> operative limits]?

This is true as far as I know. Collision diameter (propagation
delays) govern the size of an unswitched ethernet segment.
A switch isolates segments.

-- Robert
Anonymous
May 21, 2004 2:04:07 AM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

The limits you are describing below do not apply when all segments are
terminated at a switch.

Walter Roberson wrote:

> I have been faithfully evolving our LAN within the hop-count limits
> that I have read about in newsgroups and online -- namely that
> one is permitted 4 intermediate hops between end-stations for 10 Mbps,
> and 3 intermediate hops between end-stations for 100 Mbps, with
> each router "resetting" the count.

Wherever you read those you need to find better sources of information.

> Several months ago, I went to check the exact name of the "rule"
> (which I knew as the "3-4-5 rule"), and discovered that the "5"
> referred to permitted fan-outs in a star topology; and I knew from
> my readings and from practical experience that fan outs beyond 5
> are extremely common (e.g., 24 or 48 port switches, potentially
> stacked into the hundreds of ports range.)

The 5-4-3 rule applies specifically to the use of fiber-optiic links in
10base5, although it has been to some extent generalized. It does not have
direct application to any Ethernet variant that uses a star topology.

> At that point I started to doubt that the rule was for the same -kind-
> of LAN that is commonly deployed these days, and the more I thought about
> those "rules", the more I have come to doubt that what I had always
> "known" to be a hard "Thall Shalt Not" rule had any meaning for our
> network.
>
> I did some digging in the 802.3 specs last night, and I don't find
> that "rule" anywhere, and what I did find seems to lend credence
> to my thought processes. I would appreciate some confirmation or
> refutation from those who know ethernet better than I.

The "5-4-3" rule is a "rule of thumb", not part of the spec.

> The limits I found in 802.3 were on Class I and Class I repeaters
> (which I tend to think of as 'hubs'). I found a sentance in the
> 100BaseTX section that indicated that when you are using the
> classic star topology with duplex links with bridges between the
> segments, then each segment is treated as an independant collision
> domain, and the segment size limit is then bound just by the 100 metres
> limit rather than by round-trip propogation limits. But then a later
> sentance talks about limits on the "network" without any proximate
> wording to clarify if "network" is referring to a single
> collision domain.
>
>
> Am, then, I now properly interpreting that in the case where every device
> is connected to a switch port through a proper Cat5 cable of < 100 metres
> each, and the switches are themselves connected together through
> proper Cat5 < 100 metres (or through 100BaseFX or 1000BaseFX if over
> 100m), then it is permitted to have an arbitrary number of switches in the
> chain [with the higher level protocol timers then becoming the
> operative limits]?

If I understand this paragraph correctly then yes, you are now properly
interpreting.

>
> If that -is- true, that there -is- no 3 hop rule at 100BaseTX /
> 1000BaseTX that is meaningful for fully switched network,

There never was such a rule for either 100baseTX or 1000BaseT (note--there
_is_ no 1000BaseTX--there was an attempt by the cabling industry to create
such a specification and in fact it was published by the EIA/TIA I
understand, but it is not part of the IEEE Ethernet standards). The rule
for 100baseTX was two segments between endpoints, 3 with a class II
repeater (IIRC--I may have Class I and Class II confused). For 1000BaseT
it is two segments between endpoints when connected by a repeater of any
kind--since nobody is maing 1000BaseT repeaters, that point is moot.

> then
> certain topological changes I have simmering would be much simplified.
> I'm planning to migrate some equipment to a new gigabit switch that
> doesn't stack with our existing multi-vlan'd 100 Mb swithces; if there
> is no hop count limit, then the new device becomes simple to introduce,
> and transition to; if there -is- a maximum of 3 switches before hitting
> a router, then I'll have to do noticable internal renumbering to
> ensure that no -actual- path exceeds 3 switches before routing.

There is no hop count limit (at least none that you are going to encounter
in the real world).

You might want to take a look at
<http://www.ethermanage.com/ethernet/ethernet.html&gt; and pick up a copy of
"Ethernet: The Definitive Guide", which covers the territory in a much
more digestible form than the standards.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)
Anonymous
May 21, 2004 2:04:08 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hello, J.!
You wrote on Thu, 20 May 2004 22:04:07 -0400:

JC> The 5-4-3 rule applies specifically to the use of fiber-optiic
JC> links in 10base5, although it has been to some extent
JC> generalized.

Fiber-optic links in 10Base5? Wow!

With best regards,
Andrey.
Anonymous
May 21, 2004 4:14:34 AM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In article <c8jgs6$nq9$1@news.aha.ru>, Andrey Tarasov <andyvt@email.com> wrote:
: WR> If that -is- true, that there -is- no 3 hop rule at 100BaseTX
: WR> /
: WR> 1000BaseTX that is meaningful for fully switched network, then

:D efault STP timers (802.1d) are based on assumption that the network diameter is
:7 hops max.

Ah, good point. Still, for my purposes, 7 hops would be enough.


: WR> I'm planning to migrate some equipment to a new gigabit switch

:Is this new gigabit switch has a layer 3 functionality? Modern Cisco switches -
:3550, 3750, 4500 (with some line cards), 6500 (Sup 720 with some line cards) -
:has non-blocking backplane and can do wirespeed routing. It would be worthwile
:to check if you can use it as a core/distribution switch.

My plan is the WS-C3750-48TS-E (or perhaps the 24 port equivilent.)
--
Oh, to be a Blobel!
Anonymous
May 21, 2004 4:14:35 AM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

Hello, Walter!
You wrote on 21 May 2004 00:14:34 GMT:

WR> My plan is the WS-C3750-48TS-E (or perhaps the 24 port
WR> equivilent.)

You know that there is only 4 (2 on 24 ports model) fiber GigE ports, don't you?
How many fiber links do you have in your existing network?

Nevertheless, for a small-to-medium sized network it's a pretty good choice for
the role of core switch. Whatever you are using now for inter-VLAN routing
(unless it's 4500 or 6500 series :-) - this switch is an excellent replacement.

With best regards,
Andrey.
Anonymous
May 21, 2004 4:18:29 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Andrey Tarasov wrote:

> Hello, J.!
> You wrote on Thu, 20 May 2004 22:04:07 -0400:
>
> JC> The 5-4-3 rule applies specifically to the use of fiber-optiic
> JC> links in 10base5, although it has been to some extent
> JC> generalized.
>
> Fiber-optic links in 10Base5? Wow!

Yup. Contrary to common belief, there's nothing particularly new about
fiber. Look up "FOIRL" or "Fiber Optic Inter Repeater Link". That's what
two of the links had to be leaving the other 3 to have devices other than
repeaters attached, hence the 5-4-3 rule--in any path through the network
there can be at most 5 segments connected by 4 repeaters, and of the 5, at
most 3 can be copper segments with devices other than repeaters attached
("populated" segments or "mixing" segments), with the other two being
FOIRLs.

> With best regards,
> Andrey.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)
Anonymous
May 21, 2004 5:41:10 AM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In article <c8iqqk$57c$1@canopus.cc.umanitoba.ca>, roberson@ibd.nrc-
cnrc.gc.ca says...
> I have been faithfully evolving our LAN within the hop-count limits
> that I have read about in newsgroups and online -- namely that
> one is permitted 4 intermediate hops between end-stations for 10 Mbps,
> and 3 intermediate hops between end-stations for 100 Mbps, with
> each router "resetting" the count.
>
> Several months ago, I went to check the exact name of the "rule"
> (which I knew as the "3-4-5 rule"), and discovered that the "5"
> referred to permitted fan-outs in a star topology; and I knew from
> my readings and from practical experience that fan outs beyond 5
> are extremely common (e.g., 24 or 48 port switches, potentially
> stacked into the hundreds of ports range.)
[snip]

Ah...where's Kevin Oberman when you need him!

The 5-4-3 rule doesn't apply to switched (well, not even to 10Base-T)
networks. There are no mixing segments, FOIRL, nor do you run in HD.
The only thing you should worry about is the convergence time for
spanning tree. The original spec called for 7 bridges. I think it's
safe to assume that with today's switches (and the beefed up switches),
even that rule becomes a moot point. But I've never been on a network
that required more than 7 bridge hops!

--

hsb

"Somehow I imagined this experience would be more rewarding" Calvin
*************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
********************************************************************
Due to the volume of email that I receive, I may not not be able to
reply to emails sent to my account. Please post a followup instead.
********************************************************************
Anonymous
May 21, 2004 7:40:21 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <c8jrj9$13tp$1@news.aha.ru>, andyvt@email.com says...
> Fiber-optic links in 10Base5? Wow!

He meant the use of FOIRL in coax based networks.


--

hsb

"Somehow I imagined this experience would be more rewarding" Calvin
*************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
********************************************************************
Due to the volume of email that I receive, I may not not be able to
reply to emails sent to my account. Please post a followup instead.
********************************************************************
Anonymous
May 21, 2004 9:50:15 AM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

Hansang Bae <uonr@alp.ee.pbz> wrote
> In article <c8iqqk$57c$1@canopus.cc.umanitoba.ca>, roberson@ibd.nrc-
> cnrc.gc.ca says...
> > I have been faithfully evolving our LAN within the hop-count limits
> > that I have read about in newsgroups and online -- namely that
> > one is permitted 4 intermediate hops between end-stations for 10 Mbps,
> > and 3 intermediate hops between end-stations for 100 Mbps, with
> > each router "resetting" the count.
> >
> > Several months ago, I went to check the exact name of the "rule"
> > (which I knew as the "3-4-5 rule"), and discovered that the "5"
> > referred to permitted fan-outs in a star topology; and I knew from
> > my readings and from practical experience that fan outs beyond 5
> > are extremely common (e.g., 24 or 48 port switches, potentially
> > stacked into the hundreds of ports range.)
> [snip]
>
> Ah...where's Kevin Oberman when you need him!
>
> The 5-4-3 rule doesn't apply to switched (well, not even to 10Base-T)
> networks. There are no mixing segments, FOIRL, nor do you run in HD.
> The only thing you should worry about is the convergence time for
> spanning tree. The original spec called for 7 bridges. I think it's
> safe to assume that with today's switches (and the beefed up switches),
> even that rule becomes a moot point. But I've never been on a network
> that required more than 7 bridge hops!

The hop limits are not mythical, except perhaps in the
mind of the OP:-)))

This is pretty much beaten to death now however there are
some further points that I think are worth mentioning.

1.
The old 5-4-3 rule was applied to 10 Mbps Layer 1 repeaters (hubs)
and was in place I believe since the early repeaters ate the
preamble bits. i.e. X preamble bits arrive at repeater and X-n
preamble bits leave the repeater. Such a network can be thought
of as a single collision domain.

A different rule limited the diameter of a 100M layer 1
collision domain to 1 or 2 devices (depending on something
or other) which must be no more than
5 meters apart. This one turned out to be pretty irrelevant
since AFIK very few 100M repeaters have been made.
I seem to recall seeing one once, no, maybe I actually touched it:) 

GBE also included repeaters and collisions and all that stuff
but I have never seen one of those and don't expect to either.
I don't know or care what the rules are.

2.
802.1d *recommends* that the diameter of a broadcast domain
should be no more than 7 bridges (switches). It seems that this
is due to Spanning Tree issues.

I have 802.1d (1998) and it includes a section "Calculating spanning
tree parameters" so that you can roll your own. I should warn you
that it extends to 9 pages and contains a LOT of arithmetic:-)

Remember what the function of STP is. If you mess with it and
break it you could get network loops. Always consider
the worst case. e.g. Nework full of traffic with
full output queues on devices resulting in maximum traffic
delays.

3.
I am aware of NO fan out limitations other than the limits
of the forwarding table size and learning rate which vary from
product to product, and the amount of broadcast
traffic that can be tolerated. (I once did troubleshoot a new
network that was not working as expected since it turned out
the backbone links were constructed using bridge/switch
devices (Bitches?) that could only handle 4 mac addresses in
the forwarding table.)

Most of the cisco stuff has high limits on this although the
original 4000/2948G family was limited in learning rate.

4.
Remember that most modern switches (and ANY that do forwarding
between ports of differing speeds e.g. 10M to 100M) are store
and forward devices. Depending on the network speed you are
using and the behaviour of the end devices, more hops means
less throughput except in exceptional cases. e.g. Efficient
sliding window protocols applied to suitable traffic
profiles.

5.
There is little reason and I think that it bad practise
to have large diameter L2 networks, there is just no need.
Do fan out, not long strings with many hops.

6.
A thought(ette), If you don't have loops maybe the 802.1d limit
does not apply at all. I can't see why it should. Maybe if you
have a core with loops and loop free branches you could limit
the loops-present core to 7 hops and have loop free extensions
out of that region? I wouldn't like it but as always cost
considerations might make it a viable solution one day.

7.
Each additional hop would increase the probability of frame
corruption. This is due to the risk of the bridge corrupting
the frame.
Anonymous
May 21, 2004 12:02:20 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

Walter Roberson wrote:

> I have been faithfully evolving our LAN within the hop-count limits
> that I have read about in newsgroups and online -- namely that
> one is permitted 4 intermediate hops between end-stations for 10 Mbps,
> and 3 intermediate hops between end-stations for 100 Mbps, with
> each router "resetting" the count.

For 10baseT it is actually five repeaters and six segments.
(Collision detect is slightly faster.) The difference
between four and five means that you can have a central
repeater with repeaters connected to each port, with
repeaters connected to each port of those.

But these days, the central hub (not a generic term
for repeater) should be a switch.

> Several months ago, I went to check the exact name of the "rule"
> (which I knew as the "3-4-5 rule"), and discovered that the "5"
> referred to permitted fan-outs in a star topology; and I knew from
> my readings and from practical experience that fan outs beyond 5
> are extremely common (e.g., 24 or 48 port switches, potentially
> stacked into the hundreds of ports range.)

A 24 port repeater should count as one hop between any
pair of ports. Some allow stacking through a special cable,
and in that case the whole stack sometimes counts as two
hops.

> At that point I started to doubt that the rule was for the same -kind-
> of LAN that is commonly deployed these days, and the more I thought about
> those "rules", the more I have come to doubt that what I had always
> "known" to be a hard "Thall Shalt Not" rule had any meaning for our
> network.

The 5-4-3 rule is pretty useless in UTP networks. It is
from the coax and FOIRL days.

> I did some digging in the 802.3 specs last night, and I don't find
> that "rule" anywhere, and what I did find seems to lend credence
> to my thought processes. I would appreciate some confirmation or
> refutation from those who know ethernet better than I.

The rules are called model 1 and model 2, with 5-4-3 being
a convenient simplification from days gone by.

> The limits I found in 802.3 were on Class I and Class I repeaters
> (which I tend to think of as 'hubs'). I found a sentance in the
> 100BaseTX section that indicated that when you are using the
> classic star topology with duplex links with bridges between the
> segments, then each segment is treated as an independant collision
> domain, and the segment size limit is then bound just by the 100 metres
> limit rather than by round-trip propogation limits. But then a later
> sentance talks about limits on the "network" without any proximate
> wording to clarify if "network" is referring to a single
> collision domain.

For 100baseT repeaters, there are class I and class II, with
slightly different limits between them. Best choice these
days is to use a central switch, possibly connected to nearby
100baseT repeaters or dual speed repeaters, then you should
not have to worry about hop counts.

> Am, then, I now properly interpreting that in the case where every device
> is connected to a switch port through a proper Cat5 cable of < 100 metres
> each, and the switches are themselves connected together through
> proper Cat5 < 100 metres (or through 100BaseFX or 1000BaseFX if over 100m),
> then it is permitted to have an arbitrary number of switches in the
> chain [with the higher level protocol timers then becoming the
> operative limits]?

There are some limits based on convergence times for spanning
tree, though for most sized networks I wouldn't expect you to
run into them.

> If that -is- true, that there -is- no 3 hop rule at 100BaseTX /
> 1000BaseTX that is meaningful for fully switched network, then
> certain topological changes I have simmering would be much simplified.
> I'm planning to migrate some equipment to a new gigabit switch that
> doesn't stack with our existing multi-vlan'd 100 Mb swithces; if there
> is no hop count limit, then the new device becomes simple to introduce,
> and transition to; if there -is- a maximum of 3 switches before hitting
> a router, then I'll have to do noticable internal renumbering to
> ensure that no -actual- path exceeds 3 switches before routing.

As far as I know, three should be just fine. The number
seven seems to be floating around.

-- glen
Anonymous
May 21, 2004 7:34:04 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

"J. Clarke" <jclarke@nospam.invalid> wrote in message news:<c8k0kg024bo@news1.newsguy.com>...
> Andrey Tarasov wrote:
>
> > Hello, J.!
> > You wrote on Thu, 20 May 2004 22:04:07 -0400:
> >
> > JC> The 5-4-3 rule applies specifically to the use of fiber-optiic
> > JC> links in 10base5, although it has been to some extent
> > JC> generalized.
> >
> > Fiber-optic links in 10Base5? Wow!
>
> Yup. Contrary to common belief, there's nothing particularly new about
> fiber. Look up "FOIRL" or "Fiber Optic Inter Repeater Link". That's what
> two of the links had to be leaving the other 3 to have devices other than
> repeaters attached, hence the 5-4-3 rule--in any path through the network
> there can be at most 5 segments connected by 4 repeaters, and of the 5, at
> most 3 can be copper segments with devices other than repeaters attached
> ("populated" segments or "mixing" segments), with the other two being
> FOIRLs.
>
> > With best regards,
> > Andrey.

AFAIK the two unpopulated segments didn't HAVE to be fibre (they could
be copper), they simply had to be unpopulated and only used to extend
the distance of the collision domain, but I'm prepared to be corrected
on that one.

The reason for the maximum distance was that the overall length should
not be more than the distance that 1/2 a minimum size Ethernet frame
could travel. This was so that a device at the very end of the
segment, that transmitted a 64 byte frame, would be able to detect a
collision with a device at the far end of the segment (if it had sent
more than 32 bytes before the collision occurred, it would have
finished sending the frame by the time the collision made its way back
the the device's transceiver and would therefore miss the fact that a
collision had occurred). It should also detect a collision within the
first 64 bytes of a larger frame. Hence a late collision would occur
if the collision was detected after the first 64 bytes of the larger
frame.

Because a bridge was originally a store and forward device, the count
was reset as soon as one was introduced into the network. Therefore a
switch falls into the same category and resets the count.

That's my understanding anyway.

Pete
Anonymous
May 21, 2004 8:11:13 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In article <c8jq92$12g4$1@news.aha.ru>,
Andrey Tarasov <andyvt@email.com> wrote:
: WR> My plan is the WS-C3750-48TS-E (or perhaps the 24 port
: WR> equivilent.)

:You know that there is only 4 (2 on 24 ports model) fiber GigE ports, don't you?

Yes.

:How many fiber links do you have in your existing network?

We only have 2 internal gig fibre links at the moment. We're planning
on placing the device in a location where we can use cat5 to connect
to the servers that need the gig ports, and we're planning on
short-circuiting those existing gig fibre runs. We would end up
using only one of the SFP ports for now. We can't justify upgrading
the rest of the building -this- fiscal year. We have a new building
coming live -next- fiscal year, and we'll handle that by placing
in a different 3750 model such as a C3750G-12 to handle the gig fibre
concentration.

Some of our wiring closets could be handled entirely through a
10 Mb connection. Most parts of our building are not bandwidth
hungr; we probably have more use in one particular wing than
we have in the rest of the building combined -- so we'll upgrade
the hotspots in that wing and wait for demand on the others to
catch up before we spend the money. When gig comes to the other wings,
it will likely be a more political decision than technical.
--
And the wind keeps blowing the angel / Backwards into the future /
And this wind, this wind / Is called / Progress.
-- Laurie Anderson
Anonymous
May 21, 2004 8:38:22 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <c8jpk4031g9@news3.newsguy.com>,
J. Clarke <jclarke@nospam.invalid> wrote:
|Walter Roberson wrote:

|> I have been faithfully evolving our LAN within the hop-count limits
|> that I have read about in newsgroups and online -- namely that
|> one is permitted 4 intermediate hops between end-stations for 10 Mbps,
|> and 3 intermediate hops between end-stations for 100 Mbps, with
|> each router "resetting" the count.

|Wherever you read those you need to find better sources of information.

Apparently so ;-) I had looked some time earlier and found what
appeared to be an official IEEE spec that spelt out the timings
and hop counts in very gory detail, so I did have grounds for believing
the limits -- but what I hadn't looked at at the time was
the effect of going full duplex or of introducing bridges. Sometimes
in standards, the important thing is to notice what they do NOT say,
and I failed to notice at the time what was missing.


:There never was such a rule for either 100baseTX or 1000BaseT

Thank you. You and the others have convinced me that I do not need to
adhere to some constraints I had -thought- to be necessary. Which
will make things easier for me.
--
Entropy is the logarithm of probability -- Boltzmann
Anonymous
May 21, 2004 8:48:17 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In article <3adc58e4.0405210450.705e5a0d@posting.google.com>,
AnyBody43 <anybody43@hotmail.com> wrote:
:A different rule limited the diameter of a 100M layer 1
:collision domain to 1 or 2 devices (depending on something
:o r other) which must be no more than
:5 meters apart. This one turned out to be pretty irrelevant
:since AFIK very few 100M repeaters have been made.
:I seem to recall seeing one once, no, maybe I actually touched it:) 

We have one sitting around. When we bought it, we -thought- we
were getting a 10/100 switch. What it turned out to be, though,
was a 10 Mb hub and a 100 Mb hub combined, with a bridge between
the two (and automatic selection for any one port as to which
of the hubs it would participate in.) Oh, and the bridge part cost
extra: if you just had the base unit, any 10 Mb device would talk
to any other 10 Mb device, and any 100 Mb device would talk to
any ohter 100 Mb device, but the two speeds couldn't talk together.



:I am aware of NO fan out limitations other than the limits
:o f the forwarding table size and learning rate which vary from
:p roduct to product, and the amount of broadcast
:traffic that can be tolerated.

That reminds me of the practical point that if one has active NETBIOS
then locking of resources is handled by the requestor advertising
that it has the resource, and if it does not hear back a
"No you don't, I already own it!" packet within a certain timeframe,
then it is allowed to go ahead and use the resource. That has
implications about the maximum network delay when NETBIOS is involved.
--
Studies show that the average reader ignores 106% of all statistics
they see in .signatures.
Anonymous
May 21, 2004 10:27:16 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

AnyBody43 wrote:

(snip)

> The hop limits are not mythical, except perhaps in the
> mind of the OP:-)))

> This is pretty much beaten to death now however there are
> some further points that I think are worth mentioning.

> 1.
> The old 5-4-3 rule was applied to 10 Mbps Layer 1 repeaters (hubs)
> and was in place I believe since the early repeaters ate the
> preamble bits. i.e. X preamble bits arrive at repeater and X-n
> preamble bits leave the repeater. Such a network can be thought
> of as a single collision domain.

For 10Mb/s the other limit comes from IPG (inter packet gap)
loss. I believe it is that the repeater regenerates the
preamble, but that the time between packets then get
slightly shorter. In any case, it is IPG loss that limits
10baseT to five repeaters (and six segments even of 150m),
unless you add too many 50m transceiver cables.

> A different rule limited the diameter of a 100M layer 1
> collision domain to 1 or 2 devices (depending on something
> or other) which must be no more than
> 5 meters apart. This one turned out to be pretty irrelevant
> since AFIK very few 100M repeaters have been made.
> I seem to recall seeing one once, no, maybe I actually touched it:) 

It isn't really that it is 5m between them, but that the
total is 205m. Because the structured wiring standards allow
100m cables, the allowance is made for that, leaving 5m.

If the cables are short enough, the model 2 rules allow
three repeaters for 100baseTX.

I once had a 3C250, one of the early 100baseTX repeaters,
otherwise dual speed repeaters (two repeaters and a
two port bridge between them, where ports automatically
select the speed) were not so uncommon.

> GBE also included repeaters and collisions and all that stuff
> but I have never seen one of those and don't expect to either.
> I don't know or care what the rules are.

I have seen 8 port gigabit switches advertised down to $80.
At prices like that gigabit repeaters will never appear.


(snip discussion of spanning tree in switched networks)

-- glen
Anonymous
May 22, 2004 3:23:03 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Pete Mainwaring wrote:
(snip)

> AFAIK the two unpopulated segments didn't HAVE to be fibre (they could
> be copper), they simply had to be unpopulated and only used to extend
> the distance of the collision domain, but I'm prepared to be corrected
> on that one.

They can be copper, but not coaxial. Collision detect
is slower on coaxial ethernet.

> The reason for the maximum distance was that the overall length should
> not be more than the distance that 1/2 a minimum size Ethernet frame
> could travel. This was so that a device at the very end of the
> segment, that transmitted a 64 byte frame, would be able to detect a
> collision with a device at the far end of the segment (if it had sent
> more than 32 bytes before the collision occurred, it would have
> finished sending the frame by the time the collision made its way back
> the the device's transceiver and would therefore miss the fact that a
> collision had occurred). It should also detect a collision within the
> first 64 bytes of a larger frame. Hence a late collision would occur
> if the collision was detected after the first 64 bytes of the larger
> frame.

Well, it is the round trip time for a whole packet. It is
not only propagation, but there are delays in repeaters,
including collision detect. It must be that notice of a collision
reaches the sender before the packet has finished being sent.
That is why collision detect time is part of the calculation.

However, for 10baseT, assuming no transceiver cables,
six 150m segments and five repeaters are not very close to
the time limit, but more than five repeaters will exceed
the inter packet gap loss limit. That is the reason the
3-4-5 rule doesn't apply.

Also, for 10baseFL you can reach the time limit with two
segments and one repeater.

> Because a bridge was originally a store and forward device, the count
> was reset as soon as one was introduced into the network. Therefore a
> switch falls into the same category and resets the count.

Well, it is also true for bridges that start sending before
the entire packet has been received. Well, they must buffer
it anyway, in case of collisions.

-- glen
Anonymous
May 22, 2004 4:36:13 AM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <64d77c87.0405211434.7c581d16@posting.google.com>,
peter.mainwaring@virgin.net says...
> AFAIK the two unpopulated segments didn't HAVE to be fibre (they could
> be copper), they simply had to be unpopulated and only used to extend
> the distance of the collision domain, but I'm prepared to be corrected
> on that one.

Not unpopulated. That's another myth. It just couldn't be coax since
collision detection was slower on coax.


--

hsb

"Somehow I imagined this experience would be more rewarding" Calvin
*************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
********************************************************************
Due to the volume of email that I receive, I may not not be able to
reply to emails sent to my account. Please post a followup instead.
********************************************************************
Anonymous
May 22, 2004 12:27:11 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

In article <c8jgs6$nq9$1@news.aha.ru>, Andrey Tarasov <andyvt@email.com> wrote:
>Default STP timers (802.1d) are based on assumption that the network
>diameter is 7 hops max.

While this maybe true for 10 or 100Mbps networks, 1Gbps switches have
better STP standards to follow. If you keep the core (the part that
loops) at 1Gbps, you will run out of ideas and money before hitting
the limit (the STP cost limit is around 50 million hops, order of
magnitude). Before this is hit, you'll run out of time for STP I
suspect, which means, if you can tolerate the network latencies, you
can build it. The upper limit is in the 5-15 second range. At the
edge with the slower switches, don't have loops. Doing it this
way, you won't ever face a hop limit.

If you want loops and 10 or 100 in the core, don't go beyond 7, sorry.
Spend the $80 for that 8 port gige switch.
Anonymous
May 22, 2004 3:02:27 PM

Archived from groups: comp.dcom.lans.ethernet,comp.dcom.sys.cisco (More info?)

Hello, Mike!
You wrote on Sat, 22 May 2004 08:27:11 GMT:

>> Default STP timers (802.1d) are based on assumption that the
>> network diameter is 7 hops max.

MS> While this maybe true for 10 or 100Mbps networks, 1Gbps
MS> switches have better STP standards to follow.

Example? That's the first time I hear about special STP standard for 1Gbps
switches.

MS> If you want loops and 10 or 100 in the core, don't go beyond
MS> 7, sorry.
MS> Spend the $80 for that 8 port gige switch.

Sorry, I don't believe in GigE switches which cost $10 per port.

With best regards,
Andrey.
!