Cisco Catalyst 3750G StackWise Technology

[jplenbrook]

Hi everybody,

First of all, I am very excited to be able to join Cisco Support Community forum! Great stuff. I hope I can contribute as well in the very near future.

Now, we are looking into purchasing three WS-C3750G-48TS-S and stacking all of them together.

Now, as I am fresh to this technology I have some questions.

1. If you stack three switches with the StackWise cable, do you need to connect them with Ethernet Cable as well? In my understanding, connecting three switches with the StackWise cable configures them to act as a one physical device just like chassis based switches Catalyst 4500 or 6500.

Let's say we have multiple VLANs, do I need to configure VTP and trunks between them? Or would the StackWise cable just configure all of it as one switch so I don't need trunk between the switches? I know this may be a stupid question, but I am not too sure about that.

2. I also understand that there are multiple ways of connecting your switches together with the StackWise cable, however the most redundant connection would be so that the top switch connects with the bottom one, and the switches in between interconnect to the top and to the bottom with each other. Is that correct?

I have attached a screenshot from Cisco website, would this be the most efficient configuration?

3. What happens when the master switch (which has got connectivity to the router) goes offline? I understand that the new master election will happen based on priority or MAC, however new master switch would not have a link to the router and the outside. Is there any redundant configuration so let's say when master goes down, I can just unplug the cable and connect it to the new master? We use 1941/K9 as a default gateway and we do not have switch port module installed at the moment so there is only 1 LAN port available running multiple vLANs.

Thanks guys in advance and I look forward to hearing from you.

[Reza Sharifi]

Hi and welcome to the forum!

Great place to lean and contribute.

First, if you are planning to purchase new switches, the 3750G is already end of life and support.  Instead you want to look at 3650 series or 3850 series.  I will post  links to both data sheets.

For question 1, when you stack the switches together, they logically become one switch so, there is no need to connect them via Ethernet cables.

for question 2, yes, you want to stack them just like the image you posted.  switch-1 connect to switch-2, switch-2 connects to switch-3 and switch-3 connect to switch-1

For question 3, when you connect a device (router or switch) to a stack, you want to make sure you connect that device to multiple switches in the stack and not just to one. If let say you connect a router to only one of the stack switch (master or member) and if that one device fails you will lose connectivity to the router.  So you want redundant connection from the stack to the router or switch using a Portchannel.

HTH

[jplenbrook]

Hello Reza,

Thank you for your reply.

At the moment, we are going to go with 3750G as they are way cheaper and I have been assigned a budget of £1300 for two or three switches. Unfortunately, for this amount I wouldn't be able to get one 3650 or 3850 24 port switch, therefore we are looking at 3750G. I couldn't think of any other unit within this price limit. That would be great if we could go with 3850... I totally agree.

I know that the EtherChannel would provide redundancy, however we don't have spare ports on our router. We might end up getting switch module installed as this would allow redundancy. Apart from that, can you think of any other configuration that would allow us to be able just to unplug the cable from the failed unit and connect it to another port on the next master switch to allow WAN connectivity? I wonder if having one preconfigured on each other switch, just for this sort of scenario would be the solution for the time being. So if the main switch which gives an uplink dies, we can just reconnect the cable to preconfigured interface.

Also, do you know how the configuration of the stack is being saved? Let's say we stack of three 3750G switches and then we get 144 ports which we manage through the SSH or console. I assume that they will appear like an example below when I run show interfaces command:

1/12 <- 12 port on sw1

2/12 <- 12 port on sw2

3/12 <- 12 port on sw3

Is that correct? If this is correct, I then make a change to let's say 2/12 switching this port to specific vlan. As this interface is located on sw2, would the config be saved onto NVRAM on SW2 or would there be one location where the config for whole stack goes?

Do you think that there is a lot of difference between 3750G and 2960G? Is it worth getting 3750G for client access or does it make more sense to get 2960G, however they would come with no stack modules which is a bit of pain.

[Reza Sharifi · Accepted solution]

Hi,

I completely understand the budget limitation.  In this case, I would say go with 3750G vs 2960 series.  The 3750 series switches are much more feature rich than the 2960 series.  Also the 2960 series switches are mainly used for layer-2 with no stacking unless you use 2960s or 2960X, which I would not recommend.

So, lets answer your question;

As you already noted, in worse case scenario you can always configure a second port ready just in case the primary link fails. I am assuming the router is connected to the service provider for Internet access.  If this is the case and the internet provider handing off an Ethernet connection to you, you can always use a switch facing the provider.  There are some limitation in using a switch vs a router i.e switch does not support NAT, but if you simply connecting to a provider and the provider does the NAT for you than you can use a switch.  I would need to know more about your design to help in this area.

As for the stack configuration, the entire stack will act as one switch with all config being performed on the master. The master than sync the config to the members. When the master fails there will be an election and the member with highest  priority becomes the master. For management you telnet or SSH only to the master with one IP address.

Also, the port numbering will look something like this;

Switch-1 port 1 will be 1/0/1, switch-2 port 1 will show as 2/0/1 and switch-3 port 1 will show as 3/0/1 and switch-3 port 35 with show as 3/0/35 and so on. So, the first number is always the switch number than there is a 0 and than the port number.

HTH

[jplenbrook]

Hello again,

Thank you for your response. I will then purchase three 3750g for the access layer. I am just starting with Cisco Switches as we are upgrading our infrastructure from HP to Cisco, however we run Cisco routers on every single site and two MPLS'es so I don't think we will get the switch facing MPLS as we wouldn't be able to connect it to the BGP AS and they wouldn't be able to match our requirements (I thought about it earlier on). Our network is quite complicated, those switches are just for access layer, this is the first time I will be implementing stack-able Cisco switches and this is where my obvious questions came from.

You have helped me to understand this technology further and I really appreciate it.

Just one more question, do I need to get 3x0.5m stackwise cables or 2x0.5m and 1x 1m to be able to stack them? I think that I should be ok with 3x 0.5m, is that right?

[Reza Sharifi]

Hi,

You should be fine with 0.5 meter stacking cable, as long as the switches are going to be stacked right on top of one another. Also, make sure to label each switch, so you know what switch is the master and what switches are members.  For example, if you have 3 switches installed in a rack or cabinet make switch-1 (the top switch) the master and label it and the middle switch would be memeber-1 and finally the last switch be member 2.

Good Luck with your project.

Reza

[jplenbrook]

Sure thing, I couldn't live without clear labeling.

I think about setting manual priorities to each stack-able switch, for instance 15 for the master, then 14 for the next member which I will want to become master in case the primary one fails. Is that good a practice?

Like we mentioned earlier, switch ports will be labelled 1/0/x, 2/0/x and 3/0/x.

When the master switch dies with switch ports starting 1/0/x, new master election happens and switch with switch ports starting at 2/0/x is now a master switch. Do all of the port numbers move to replace the old master? Let's say new master (2/0/x) which has replaced the old one (1/0/x) will now become 1/0/x or will the port numbers stay the same (2/0/x) ?

In relation to stacking, I've read somewhere about 32 Gbps total bandwidth available when you stack the switches. Please, could you elaborate on it a bit more if you know anything about that? How does that work? Do the stack cables provide bandwidth as well? If so, is it 32 Gbps across the whole stack or in between each other switch?

Thanks for all your advise.

[Reza Sharifi]

or will the port numbers stay the same (2/0/x) ?

They will stay the same (2/0/x).

I don't remember the 3560G stack bandwidth number but if data sheet says 32Gbps than that is across the whole stack. To get the most stack bandwidth it is necessary for each switch to be connected to 2 other switches.  You can only use one stacking cable and the stack will work fine, but that means the bandwidth will be 50% of 32 (16).

Thanks,

[jplenbrook]

Hi Reza,

Thanks for that.

Cisco StackWise stacking for 3750G creates a 32-Gbps switch interconnection.

Do you think that setting switch priorities to static (15,14 and so on) like I mentioned earlier is a good practice?

Do you know if the 3750G four SFP up links are 10GBASE or 1000BASE? The data sheet is not clear on that point. It just states:

• Cisco Catalyst 3750G-48TS-48 Ethernet 10/100/1000 ports and four SFP uplinks

I just wonder if they are 10Gb (single or multi mode?), could they be used on etherchannels to improve the stack forwarding bandwidth? 2 fibres would be going to the switch above and 2 would be going to the switch below, creating 20 Gbps in between each switch. Is that how that works? 

Cisco website states:

• When uplink capacity needs to be increased, you can easily upgrade your bandwidth by adding a 10 Gigabit Ethernet version to the stack and upgrade your Gigabit Ethernet links with 10 Gigabit Ethernet on the existing fiber.

This is about the uplink capacity of the stack so I understand it as uplink to the router/ISP, not speed transfers in between the switches.

I just imagine the top bandwidth usage scenario where we have 40 devices connected to 48 port switch, each device using up to 1 Gbps. Therefore, it would be nice to have an uplink to the next switch supporting more bandwidth to allow higher throughput on the LAN.

[devils_advocate · Accepted solution]

Do you think that setting switch priorities to static (15,14 and so on) like I mentioned earlier is a good practice?

I am unsure if Cisco suggest its good practice or whether it is just good practice based on opinion from those with experience. When the stack boots it will go through an election process to select the master. If you have not set the priorities manually it may take longer for the election process to happen, hence why people set them manually. Another reason is that you know what switch will be the master, rather than relying on the switches working it out themselves. 

Do you know if the 3750G four SFP up links are 10GBASE or 1000BASE?

The built in SFP ports on the 3750G switches are gigabit, I don't believe its possible to use them for anything more than this.
SFP normally means Gigabit
SFP+ normally means Gigabit or 10Gb depending on the SFP type you have purchased.

FYI - You can use any of the ports (SFP or Copper) to create etherchannels to increase bandwidth. You could (in theory) bundle all 4x SFP ports together to form a 4Gb etherchannel.

Cisco website states:

• When uplink capacity needs to be increased, you can easily upgrade your bandwidth by adding a 10 Gigabit Ethernet version to the stack and upgrade your Gigabit Ethernet links with 10 Gigabit Ethernet on the existing fiber.

What I believe this means is that you can add one of the later 3750x series switches to the stack later on which support 10Gb uplinks. The 3750x series do not have built in SFP/SFP+ ports, you need to purchase an additional module for this.

Hope this helps

[jplenbrook]

Hi,

Thank you very much for your reply. I apperciate it.

Everything makes sense.

Do you know what is the difference between the normal gigabit ethernet port and SFP port if they are capable of running at the same speed anyway?

Thanks.

[devils_advocate]

In terms of the 3750G line of switches, I believe the only difference is difference in the medium used to connect one switch to another.

Copper has a limit of 100 metres so if you had one 3750G in one building and one 3750G in another, the chance of the run being less than 100m is slim unless the buildings are next to each other. 

Fibre (depending on the type) can run for miles so although the ports are still only Gigabit in speed, they are needed when you are uplinking/downlinking switches which are more than the distance limits of copper. 

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

As Devils Advocate has already noted, the copper ports, on the 3750G series, are copper gig.  The SPF ports, also gig, support different gig media, including copper, however they are usually used for fiber connections.  The kind of SFP transceiver determines the distance they support (and their cost).  Fiber, besides supporting (much) longer distances than copper, isn't impacted by EM interference and often has slightly better link down detection, than copper.  The latter one reason fiber will also be used for infrastructure links even when copper would work.  (I've used edge copper ports for infrastructure, without any real issues.  However, also note, on the 3750 series, buffer RAM is dedicated to banks of 24 copper ports and to the SPF ports, i.e. SFP ports have more buffer RAM to share between themselves.)

BTW, in answer to some of your earlier questions . . .

StackWise's 32 Gbps, I believe, is dual 8 Gbps (duplex) links.  StackWise floods everything onto the ring, and the sender removes traffic from the ring (i.e. not an optimal design for a really busy switch - but often fine for light usage).

There was one 3750G model that provided a single 10g port; wasn't very popular as the port could only support 8 Gbps of throughput.

The later StackWisePlus series (the E and -X), can be stacked with original series 3750s. This is another way to get 10g into an original 3750 series stack, but I would recommend against that.

[jplenbrook]

Hi

Thanks for your reply.

1. Are you saying that I would be better off with not using StackWise and connecting the switches with etherchannels instead, if our network is rather busy?

2. If we decide to go with StackWise, can we increase the bandwidth across the stack with additional etherchannels at the same time or do we have to only use one type of connection at the time; either StackWise or etherchannels? 

3. I've now watched a few YouTube documents and read some articles about how fast the links are.

 - Bidirectional connection = 16 Gbps per StackWise cable or does the StackWise cable has got 2x16 Gbps inside? I am a bit confused about how the bandwidth works across the stack. You mentioned 8 Gbps earlier...

As I have shown on a screenshot in my original post, each switch will be connected to its neighbors with 2 StackWise cables. Does that mean that each switch receive 16Gbps of channel inter-connectivity between each other, 32 Gbps in total?

Does that mean that if I have 3 switches in the stack, the total backbone speed of the stack will be 96 Gbps?

Thanks!

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

#1 Normally no, i.e. StackWise would generally be best.  However, in some really extreme cases, you might be better, from a performance standpoint, using ordinary Ethernet cables.  But remember, with StackWise you have dual 8 Gbps to start with, i.e. 16x a single gig cable.  Also, you would lose the other advantages of StackWise, i.e. those from having one logical unit.

#2 It wouldn't work.  As a 3750 stack operates as a single unit, connecting stack members would be self looping.

#3 Yes, I said 8 Gbps, duplex.  Think of a full duplex gig port, you have 2 Gbps of bandwidth, if you count both in and out.  So with StackWise, a duplex 8 Gbps provides 16 Gbps of bandwidth, and if you have two of them, you have your 32 Gbps StackWise bandwidth.

As how the a StackRing works, every stack member places all its traffic on the ring, whether it needs to go off the local stack member or not.  And, even if another switch member "needs" the traffic, it continues around the ring until the original stack member removes it.  So, effectively your StackWise ring will perform much like a 16 Gbps bus.  (IMO, should be fine with an original 3750 stack, i.e. before even the 3750G models.  However, with the 3750Gs, just one 48 port model has 52 G of interface bandwidth.  This is probably why the follow on model, the 3750E has twice the ring bandwidth, but probably just, if not more, important, the 3750E's StackWisePlus only places [unicast] traffic on the ring if it needs to, and the receiving stack member takes it off.)

To your last question, with 3 StackWise switches, you still have just 32 Gbps to share.  (NB:  With StackWisePlus, yes and no.  In theory if switch member 1 only "talked" with switch member 2 which only "talked" with switch member 3 which only "talked" with switch member 1.  In practice, you'll likely have some transit traffic.)