Cat 4000/4500/5000/6000: router or switch?

I wonder if someone could explain briefly why the Cat 4000,
4500, 5000, and 6000 are classified by Cisco as Switches instead of
Routers?

I can see the point for the 5000 line, in that the 5000
just does layer 2 unless you add in the optional RSM, and
people really do use the 5000 as a pure switch. Probably
some of the 4000 line too.

In particular why is the new 4503 classed as a switch, when
the lowest supervisor engine you can get for it (Supervisor II+)
has layer 3 routing and supports traditional router-only
features such as GRE (in software until you get to the IV engine) ?

I'm accustomed to the 3550 and 3750 series being referred to
as "multilayer switches", as they do not support software
interfaces such as loopbacks and GRE tunnels... but the 450x line
does support those.

And where does the new 4948 "multilayer switch" fit into this?
It is fixed configuration (like the 2950/3550/3750 in that
regard), but with the -E software supports OSPF, EIGRP and
some other unspecified goodies [I can't find the list of
differences.]

The feature navigator claims that the 4500 with the new
Supervisor Engine II+TS does not support GRE tunnels, but the
release (20) release notes say that all the 4500 support GRE.
The feature navigator either doesn't know about the 4948 yet or
else classifies it as being part of some other family
(e.g., the 4000)


A semi-related question: if I am looking for gigabit ports
(mostly TX but some SPF or GBIC), layer 3 routing, and QoS
with traffic control, and (not sure yet) perhaps room for
expansion into VOIP, and I do not need any non-ethernet connections
(such as T1), then what would be reason I would go for the new 3845
"router" instead of (say) the new C4503-II+TS-48GB
bundle, which is a 4503 chassis and power supply, software,
and a 48 port 10/100/1000 card that has some SPFs thrown in?
(This bundle does not appear to be documented by Cisco, but it
is listed by some stores.) ?

 

Hello, Walter!
You wrote on 22 Oct 2004 19:40:23 GMT:

WR> I wonder if someone could explain briefly why the Cat 4000,
WR> 4500, 5000, and 6000 are classified by Cisco as Switches instead
WR> of Routers?

For marketing and historical reasons?

WR> And where does the new 4948 "multilayer switch" fit into this?
WR> It is fixed configuration (like the 2950/3550/3750 in that
WR> regard), but with the -E software supports OSPF, EIGRP and
WR> some other unspecified goodies [I can't find the list of
WR> differences.]

4948 looks like Sup V in a stackable form. It's more powerful than 3750 - more
TCAM, more RAM - but has a smaller footprint than normal 4500.

WR> A semi-related question: if I am looking for gigabit ports
WR> (mostly TX but some SPF or GBIC), layer 3 routing, and QoS
WR> with traffic control, and (not sure yet) perhaps room for
WR> expansion into VOIP, and I do not need any non-ethernet
WR> connections (such as T1), then what would be reason I would go
WR> for the new 3845 "router" instead of (say) the new
WR> C4503-II+TS-48GB bundle, which is a 4503 chassis and power
WR> supply, software, and a 48 port 10/100/1000 card that has some
WR> SPFs thrown in? (This bundle does not appear to be documented by
WR> Cisco, but it is listed by some stores.) ?

One reason would be if you need PSTN gateway for VoIP and want to have
everything in one box. Than 3845 will be a better solution for a small office.
Be careful with Sup II-PLUS-TS though - it only works in 4503. 48 10/100/1000
card has it's own trick - since each slot in 4500 series has only 6Gbps
connection to Supervisor, ports on this card are heavily oversubscribed - 8:1.

With best regards,
Andrey.

 

:Be careful with Sup II-PLUS-TS though - it only works in 4503. 48 10/100/1000
:card has it's own trick - since each slot in 4500 series has only 6Gbps
:connection to Supervisor, ports on this card are heavily oversubscribed - 8:1.

That's a good piece of information to have.

According to the data sheets, the Sup II+, Sup IV, and Sup V in the 4503
are 28 Gbps and are 64 Gbps in the 4506 or 4507R; also the data sheets
indicate that the Sup II+TS in the 4503 is 64 Gbps.

http://www.cisco.com/en/US/products/hw/switches/ps4324/products_data_sheet0900aecd801792b1.html

If we ignore the Sup II+TS for the moment, and split the 21 Gbps over 3
slots, the implication would be 7 Gbps. If 6 is the maximum the bus
will put through, then the implication could just be that the switch
fabric is faster than the bus and so that the Supervisor has "spare steam".
But when I look at the 64 rating for the 6 or 7 slot units, 64/6 and
64/7 are both more than the 6 Gbps that you mention. The Sup II+ is
not supported in the 10 slot unit.. if it were, 64/10 would at least be
close to the 6 Gbps you mention. The 10 slot unit requires Sup V,
which is rated at 96 Gbps, which would still be more than 6 Gbps per slot.
It does not make sense to me that they would bother requiring a higher
speed supervisor when the lower end one already has spare capacity??

I've looked through several of the datasheets now, and I do not see 6 Gbps
per slot mentioned anywhere. Would you have a link to that figure?


Either way, the fact does still remain that with the 4503 and its
28 Gbps limit, a 48 port gig card is going to oversubscribe the fabric
by about 2:1 [unless the switching doesn't have to go through the fabric
if the destination is on the same line card... then it would be down to
traffic patterns] and the problem would only get worse if one
expanded to 96 ports (2 x 48). The 4948 I see has 96 Gbps switch
fabric, which is undersubscribed 1:2... but one could imagine
something like that it had two separate 48 Gbps fabrics, one for input
and the other for output.

 

:4948 looks like Sup V in a stackable form. It's more powerful than 3750 - more
:TCAM, more RAM - but has a smaller footprint than normal 4500.

I just had another look at the specs on the 4948; I do not see any
stackability ?


: WR> no connections (such as T1), then what would be reason I would go
: WR> for the new 3845 "router" instead of (say) the new
: WR> C4503-II+TS-48GB bundle, which is a 4503 chassis and power

:One reason would be if you need PSTN gateway for VoIP and want to have
:everything in one box. Than 3845 will be a better solution for a small office.

Looking at the 4000/4500 series cards -- is there a difference between
the PSTN gateway you mention, and the facilities provided by the VIC-2FXS
or VIC-2FXO, or VIX-2FXO-EU ?

 

Hello, Walter!
You wrote on 22 Oct 2004 21:50:49 GMT:

WR> According to the data sheets, the Sup II+, Sup IV, and Sup V in
WR> the 4503 are 28 Gbps and are 64 Gbps in the 4506 or 4507R; also
WR> the data sheets indicate that the Sup II+TS in the 4503 is 64
WR> Gbps.

And that goes like this - (6Gbps x 2 slots + 1Gbs x 2 ports on Sup) x 2 for
full-duplex = 28Gbps in case of 4503. For 4506 and 4507R it would be (6 x 5 + 1
x 2) x 2 = 64Gbps.

WR> Gbps you mention. The 10 slot unit requires Sup V, which is rated
WR> at 96 Gbps, which would still be more than 6 Gbps per slot. It
WR> does not make sense to me that they would bother requiring a
WR> higher speed supervisor when the lower end one already has spare
WR> capacity??

With Sup V it's a little bit different - 8 slots x 6Gbps + 4 ports x 1Gbps
(because you can have 2 Sups and both ports will be active, which is not the
case for Sup IV) should give us 104Gbps but there is only 96. And I just found
an explanation how Cisco pulls it off -
....
*When Supervisor V is used in the 4510R chassis, Slot 10 (Flex-slot) will
support a sub-set of linecards: 2-port GBIC (WS-X4302-GB) and Access Gateway
Module (WS-X4604-GWY). Future Supervisor Engines will allow Slot 10 to
accommodate any and all linecards.
....

So correct formula is (7 x 6 + 4 x 1 + 2 x 1) x 2 = 96.

WR> I've looked through several of the datasheets now, and I do not
WR> see 6 Gbps per slot mentioned anywhere. Would you have a link to
WR> that figure?

It's in description of line cards and configuration guide for Etherchannel -

....
The Catalyst 4000/4500 Series 6-port Gigabit Ethernet line card provides six
ports of dedicated 1000BASE-X Gigabit Ethernet uplinks for high-speed backbone
....
The Cisco Catalyst 4000/4500 Series 18-Port Gigabit Ethernet line card provides
two ports of dedicated 1000BASE-X GBIC-based Gigabit Ethernet uplinks, as well
as up to 16 GBIC-based ports for high-performance Gigabit Ethernet server
connectivity. The 16 server ports can burst to Gigabit Ethernet line rate and
share 8 Gbps of full-duplex bandwidth into the switching fabric.
....
The Cisco Catalyst 4000/4500 Series 48-port 10/100/1000BASE-T line card ... The
48 ports can burst to Gigabit Ethernet line rate and share 6 Gbps of full-duplex
bandwidth into the switching fabric. The amount of oversubscription can be
controlled by varying the number of ports used at 1000 Mbps.
....

WR> only get worse if one expanded to 96 ports (2 x 48). The 4948 I
WR> see has 96 Gbps switch fabric, which is undersubscribed 1:2...
WR> but one could imagine something like that it had two separate 48
WR> Gbps fabrics, one for input and the other for output.

No, 4948 is indeed has 96Gbps fabric. 48 ports x 1Gbps x 2 for full-duplex.

With best regards,
Andrey.

 

Hello, Walter!
You wrote on 22 Oct 2004 21:59:04 GMT:

WR> I just had another look at the specs on the 4948; I do not see
WR> any stackability ?

I don't think it will provide one. You still can daisy-chain them via regular
GigE ports.

WR> Looking at the 4000/4500 series cards -- is there a difference
WR> between the PSTN gateway you mention, and the facilities provided
WR> by the VIC-2FXS or VIC-2FXO, or VIX-2FXO-EU ?

Well, I had something along T1 or PRI card in mind. There is AGM module for 4500
where you can plug those (and FXO/FXS cards also) but I don't know much about
it.

With best regards,
Andrey.

 

I received the "8 Gbps" story for this card from a supplier but
I did not like the looks of it at all. I assumed that it was an
error since I suspect that "since each slot in 4500 series
has only 6Gbps" is true.

We have 18 ports

2 x 1G dedicated
16 x 1G per block of 4 ports

= 6G and not 8G.

Just a thought.

As regards the subject line, since neither are precisely defined
it is difficult to make much of it at all. Just go with the marketing
flow, read the specs, get in tow with someone who gets access to the NDA
material and keeps up to date.

Cisco for example used to (and may well still do) give their Gold
partners consutlants detailed technical briefings on the inner
workings of all of the kit. The material was often not available
in written form The problem with this approach was that unless you
are spending quite a bit it is difficult to get access to the material
at all since the only people who have access to it are talking to
customers that are spending millions.

 

Hello, AnyBody43!
You wrote on 23 Oct 2004 13:24:43 -0700:

??>> The Cisco Catalyst 4000/4500 Series 18-Port Gigabit Ethernet
??>> line card provides two ports of dedicated 1000BASE-X GBIC-based
??>> Gigabit Ethernet uplinks, as well as up to 16 GBIC-based ports
??>> for high-performance Gigabit Ethernet server connectivity. The
??>> 16 server ports can burst to Gigabit Ethernet line rate and
??>> share 8 Gbps of full-duplex bandwidth into the switching fabric.

A> I received the "8 Gbps" story for this card from a supplier but
A> I did not like the looks of it at all. I assumed that it was an
A> error since I suspect that "since each slot in 4500 series
A> has only 6Gbps" is true.

A> We have 18 ports

A> 2 x 1G dedicated
A> 16 x 1G per block of 4 ports

A> = 6G and not 8G.

A> Just a thought.

8Gbps _full-duplex_ So it's 4Gbps connection shared by 1 ports. 4:1
oversubscription.

With best regards,
Andrey.