I'm in a purchase project of a new Gigabit switch for my company.
Currently my company has a "Catalyst 3560G-48TS". It has four additional ports called SFP that I supposed can be used to connect to another switch to constitute a sort of "backbone", right? I want to connect the current switch to the new one with as much transfer rate as possible. Ideally at least 5Gbps (yeah, if I use a single 1gbps cable, it's certainly not enough). Both switches will be separated at a distance of about 15m.
I know nothing about this SFP. Could someone tell me what are supported or supporting this? What cables should I buy? Optical fibers? What's the transfer rate could I expect from EACH SFP port?
Of course, I'm more interested to not so expensive solution, or at least a good quality/price ratio solution.
Thanks in advance.
You can put different modules into these SFP ports, like 1Gb cat5 copper or different fiber (SX,LX,ZX) connections.
Scroll to the bottom and look at the SFP modules, those are all the modules that are supported for the SFP ports on the 3560G.
You can also use the interconnect cable to link the 3560g switches.
Your not going to get faster than 1Gbps per port, because hey they are 1000 speed ports :)
You can aggregate those SFP ports and run a Layer2 or layer3 etherchannel to another switch.
Thank you for your reply.
I see. So actually it's not interesting to link switches using the SFP ports (because of the short distance between switches).
I'm wondering if this switch of mine (3560G) is compatible to port aggregation standard (IEEE 802.3ad, right?) and thus compatible to switches of other brand marks.
"I'm wondering if this switch of mine (3560G) is compatible to port aggregation standard (IEEE 802.3ad, right?) and thus compatible to switches of other brand marks."
Yes. I will hazard a guess that ALL Cisco switches now support 802.3ad LACP. In fact, if memory serves, Cisco recommends using the standards-based LACP over its own proprietary PAgP.
Be mindful that 3560s now come in 3560-E version that has 2 X2 ports. X2 ports will allow you to plug X2 SFP (i.e. 10 Gigabit Ethernet). OK, it's not exactly cheap solution. Even though you're not going to use the 10 Gigabit option, it's still better to procure 3560-E instead of the "vanilla" 3560G, because the "E" models have something like 4x the switch fabric over the "vanilla" models. Refer to this "cheat sheet" for switching performance comparison: http://www.cisco.com/web/partners/downloads/765/tools/quickreference/switchperformance.pdf
Also of note, if you need multiple switches, you might want to consider 3750/3750E that can be "stacked", which makes multiple switches appear as a single logical switch.
"Stacked" switches mean multiple switches inter-connected with "special" connections that make multiple switches to appear as a single logical unit. Single logical unit here means that the stacked switches share the same MAC address, Mgmt IP address, etc, making management easier, as well as negating the need for "first hop redundancy" (e.g. HSRP/VRRP) or reliance on Spanning Tree. All switches participating in the "stack" essentially share the same backbone through the stacking connection.
In Cisco, the only switch model that supports stacking is Catalyst 3750/3750E. The main difference between 3750 and 3750E is the switching performance. Go to 3750's page on Cisco.com. You should be able to find a link that provides comparison between these models.
I see. When you talked about 3750/3750E as a "side note", you really excluded my 3750G.
Well, I hope you're not going to suggest me to dump my 3750G which costed me more than 5000 bucks ....
Dude... 3750G is just a variety of 3750. I made a distinction of 3750E from the "normal" 3750/3750G because the "E" is quite different from the "normal" ones.
So, straight answer to your original post: the quickest possible way to connect your 3750Gs is using the StackWise cable which should be included with your switches, therefore there shouldn't be any need for your to purchase more cables and/or SFPs. Unless, for whatever reason (e.g. security), you need your switches to be "physically distinct" from each other for both physical and logical separations of your servers, then you'd have to fall back to using EtherChannel.
As far as EtherChannel goes, considering your switches are so close to each other, just use Cat5e/Cat6 patch leads to connect several ports between the two switches. Don't worry about SFPs and fiber optics.
Sorry... when I posted my previous reply, I forgot that your switches will be 15 meters apart. In this case, StackWise is out the window. You're stuck with EtherChannel. Use Cat5e/Cat6 patch leads to connect up, say, 6 ports and bundle 'em up as EtherChannel.
The thing with EtherChannel is, it is most beneficial when a fairly large number of end devices are commnicating across the bundle. If only 2 servers communicating, most likely only 1 link out of the bundle will be used.
If you want as close to "true" load balancing as possible, the best way is to configure multiple independent Layer-3 links between your switches, and either run IGP (EIGRP, OSPF, or even RIP) or create multiple static routes between them. The biggest problem with this is you're wasting a number of IP addresses, which may be OK if you use private addressing (RFC1918), but not so good if your set up involves Internet routable addressing.
The SFP ports take pluggable modules for different media, primarily different types of fiber-optic cables. There are also modules for regular copper gigabit ethernet the same as the other 48 ports, which would just give you four additional ports. In addition there is a 50cm cable available for interconnecting two switches in the same rack.
From a practical standpoint there is no difference in transfer rates using an SFP or a regular switch port. They're both gigabit ethernet. Plain old copper wire is cheapest and easiest to work with over a 15-meter distance.
Your 15 meter distance rules out the short interconnect cables.
If you have spare conventional ports you can run multiple ordinary cables to link the switches, or you can buy the modules to plug into the SFPs and run copper or fiber.
Use etherchannel (PAGP/LACP) to get higher switch-to-switch speeds than from a single cross-connect. There is a point of diminishing return, make sure that the switch fabric is capable of pushing the speed that you want.
> Plain old copper wire is cheapest and easiest to work with over a 15-meter distance.
Yes, that's also the conclusion I've got to. I've already dropped the idea to use optical fiber.
Another possibility is to move one of the switches so that they're within a meter or two of each other and use the stackwise cables to tie the switches together. you'll need to make the ethernet runs from one switch to the patch panel or hosts longer but will gain performance and not need to deal with PAGP/LACP, also free up that many ports on both switches.