Gigabit speed: Better to use three 8-port switches or one 24-port switch?

[jrichrds]

Will I get less bottleneck in speed by taking advantage of the 4 ports on my gigabit router to connect multiple 8-port gigabit switches, or am I better off going with just one 24-port gigabit switch?

Using one 24-port gigabit switch is the simpler option, but using three 8-port gigabit switches actually ends up being less expensive. I care more about speed than price though.

 

[keeleysam]

24 port will be faster, as there won't be gigabit bottlenecks between each segment.

 

[heymrdj]

I doubt your router is going to be a transfer hub, rather the switch needs to be. I'm going to make an analogy, try to follow along. Keeleysam is correct, but I like doing this too much when I'm bored .

Computers 1 and 2 on switch 1, Computers 3 and 4 on switch 2, computers 5 and 6 on switch 3.

If you use three switches this is how it'll work. Transfer between 5 and 6, full 1gbit, as they're in the same switch. Transfer between 1 and 4, 1gbit because they are using the gbit pipe between your router and the other switch. Computers 1 and 2 transfers at the same time to 4 and 6, Computers 1 and 2 could only push out at 500kb each because they have to SHARE that 1gbit pipe from switch 1 to the router, while 4 and 6 would twiddle it's thumbs, having 500kb more bandwitdth each they could be recieving.

Now using a single 24gbit is 1gbit no matter where you go. All pipes are separated, thus you have a 24 pipe, a dedicated line to each computer, rather than 3 lines feeding 24 systems.

Cliffs:

Get the 24.

 

[alpineranger]

You also don't get 24 ports for computers with three switches because you use ports to connect the three to each other. That said, there is another advantage to using the 3 switches that you might not have considered: distance. Because CSMA waits for a time period equivalent to the maximum potential propagation time of signals on the wire, they've just set a maximum distance and it's only a 100m (btw, the way the math works out, the faster you transfer signals in this scheme, if you keep the maximum frame size the same, the maximum theoretical efficiency decreases, so it's significantly more of an issue with gigabit than with fast ethernet). In the pre fast ethernet days, this would refer to your entire network segment (everything on your network of hubs), but since then, each link is for the purporses of CSMA, so you get the maximum length restriction applied only to each link, not to the whole network.

 

[jrichrds]

Thanks! I thought all data had to flow through the router to get from one computer to another, which is incorrect. So if I use one 24-port switch, the router is basically only using its link to the switch to direct traffic, right? So even if I use my existing 100mbit router, all computers connected to the gigabit switch will still transfer at gigabit speeds?

I'm going to order the Netgear JGS524 24-port gigabit switch.

 

[spidey07]

There is no good reason to use separate switches. None whatsoever.

The point about CSMA is incorrect as the interframe gap scales inversely linear with transmission speed. The IFG stays the same as a percentage of overall time on the wire relative to baudrate. The added latency of each switch means using more switches adds to overall latency, MUCH more than the interframe gap.

 

[Jeff7181]

Originally posted by: jrichrds
Thanks! I thought all data had to flow through the router to get from one computer to another, which is incorrect. So if I use one 24-port switch, the router is basically only using its link to the switch to direct traffic, right? So even if I use my existing 100mbit router, all computers connected to the gigabit switch will still transfer at gigabit speeds?

I'm going to order the Netgear JGS524 24-port gigabit switch.

The router directs traffic between different networks (subnets). If all your clients are on the same network you don't need a router to allow them to communicate.

 

[alpineranger]

The point about switching latency is well taken, as the higher the data rate, the more sensitive tcp is to latency.

Symbol rate for fast ethernet and gigabit (over utp) are the same. The interframe gap is smaller, but carrier extension is used which effectively wastes line bandwidth when you have small frames.

My point still stands though. CSMA efficiency is a function of how long you spend transmitting data in a frame vs the listening time which is set by maximum distance * signal propagation rate. Practically I would expect that since ethernet is no longer a shared medium with half-duplex links that CSMA will go away, but we still have that legacy currently.

 

[InlineFive]

Originally posted by: alpineranger
The point about switching latency is well taken, as the higher the data rate, the more sensitive tcp is to latency.

Symbol rate for fast ethernet and gigabit (over utp) are the same. The interframe gap is smaller, but carrier extension is used which effectively wastes line bandwidth when you have small frames.

My point still stands though. CSMA efficiency is a function of how long you spend transmitting data in a frame vs the listening time which is set by maximum distance * signal propagation rate. Practically I would expect that since ethernet is no longer a shared medium with half-duplex links that CSMA will go away, but we still have that legacy currently.

Based upon my limited knowledge CSMA does become a moot point on a single switch because they are all separate collision domains. However, one uplink port is one collision for the entire switch (not sure if this explanation does it justice) and thus CSMA comes into effect.