small gigE switches?

[wkearney99]

I've need to install several 5 & 8 port switches and I'm a bit puzzled about how mixes of 100 and 1000 mbps traffic is handled. I've read that a few of the lower end switches on sensing 100mbps traffic will downgrade all switch throughput to that speed. The link to/from the client devices on the ports remains gigE but actual speed through the switch drops down. Assuming a gigE uplink, of course. Sort of negates the point of having a gigE uplink or client connection if everything's going to get throttled.

So my question is has anyone experienced this and if not then what make/model switches are you using?

The scenario is multiple devices in locations not served by enough wire back to the rack for each device. As in, printers, desktops, etc. Whenever possible I prefer to keep the numbers of hops across a given network to a minimum. Specifically avoiding stuff like chaining multiple low port-count devices together, but replacing them with higher-density switches instead. But I'd prefer to avoid throttling a pair of gigE desktops just because a 100mbps printer is connected to the same switch.

That and what's the low-down on managed vs unmanaged for small port count switches? I ran across some netgear units claiming 'unmanaged+', supposedly with some rudimentary traffic prioritizing features. Any experiences pro/con with them?

 

[stevech]

I've need to install several 5 & 8 port switches and I'm a bit puzzled about how mixes of 100 and 1000 mbps traffic is handled. I've read that a few of the lower end switches on sensing 100mbps traffic will downgrade all switch throughput to that speed. The link to/from the client devices on the ports remains gigE but actual speed through the switch drops down. Assuming a gigE uplink, of course. Sort of negates the point of having a gigE uplink or client connection if everything's going to get throttled.

So my question is has anyone experienced this and if not then what make/model switches are you using?

The scenario is multiple devices in locations not served by enough wire back to the rack for each device. As in, printers, desktops, etc. Whenever possible I prefer to keep the numbers of hops across a given network to a minimum. Specifically avoiding stuff like chaining multiple low port-count devices together, but replacing them with higher-density switches instead. But I'd prefer to avoid throttling a pair of gigE desktops just because a 100mbps printer is connected to the same switch.

That and what's the low-down on managed vs unmanaged for small port count switches? I ran across some netgear units claiming 'unmanaged+', supposedly with some rudimentary traffic prioritizing features. Any experiences pro/con with them?

I haven't seen the degradation of which you speak. My home LAN has these switches
Switch built into WiFi router (Cradlepoint MBR900). It's 10/100. No fast PCs connect directly to router, of course!!!
Netgear 8 port lightly managed GS108, 10/100/1000
Fry's house brand 5 port 10/100/1000 times 2

My main PC and NAS and router are connected to the GS108. Others are for various gizmos and an added WiFi access point (ASUS NT-DR12) in same room as heavily used iPad.

Only when the NAS is copying GB files to the wired PC can I get over 100Mbps on the LAN, and that's not often. Anything WiFi here doesn't need gigE, esp. iPad and HTC One Android phone, and laptop.

 

[azazel1024]

I've seen hubs do this, I've never seen a switch do this in 15+ years of networking.

I think you'll be just fine.

 

[wkearney99]

I've seen hubs do this, I've never seen a switch do this in 15+ years of networking. I think you'll be just fine.

Sure, I'm well aware the difference between hubs and switches. I've been wrangling this nonsense since before there was an Ethernet.

I've seen plenty of low-end switches that can't keep up with wire speed. Perhaps they're a lot 'less worse' with today's gear, but then we run afoul of crappy manufacturing problems (bad caps, power supplies, etc).

When dumping a lot of video content through a switch I've generally found that 'over-subscribing' helps avoid lost traffic. As in, when you need 4 machines to be able to funnel large data then it's better to use an 8-port (or more) than to expect a 5-port to keep up. These being in-switch fabric issues, where the hosts are all sending traffic within the one switch, not across the uplink.

Couple this with modern applications that depend on avoiding latency problems (gaming, voip, video conferencing) and switch performance matters even more.

I don't want 'just fine'. I want 'known performance', especially under load.

 

[azazel1024]

Okay, sure, that may be a problem. Actually I am sure it could be. My experience is on small office and home networks where switches aren't going to be stressed that heavily.

The 8 port is a Trendnet TEG-s80g. I haven't funneled 16 gigabits of traffic through it, but it is supposedly rated for that kind of traffic (full duplex on each port with the switching fabric supporting the whole shebang), but I have run over 12 gigabits of data through it concurrently with no slow down (3+3 ports, SMB multichannel going full duplex between my desktop and my server with no noticable slow down. I was trouble shooting my main switch the other night, so I hooked the two machines up between just this switch and ran them full tilt).

Anyway, at least in my experience, I haven't see any issues with "cheap" 5 and 8 port gigabit switches of the last 4-5 years handling at least several gigabits of data concurrently or causing any single port speed issues.

My 8 port trendnet pushing the same 350MB/sec full duplex on 3+3 ports that my 16 port TP-Link SG2216 can do. I can't swear to that on a tp-link (or it may be trendnet) 5 port switch I installed, but it could at least push single port full duplex at the same 117MB/sec either of the other switches could do. I don't have access to that later switch anymore as I installed it in a rental property when I moved to leave the renters with a switch connecting the 5 lan drops in the townhouse, but I did test it briefly before installing it.

I've used the same Trendnet 8 port model for a couple of friend's home LANs and a couple of small business setups. Its cheap, its super low power and none have been defective

I am sure there are faulty switches and there might be some who's switching fabric really can't keep up, but again, I haven't seen any.

I also haven't seen any latency issues under load on these switches (as well as a couple of different netgear 5/8/16 port models).

I have never seen a switch push all of the ports or traffic down to a slower speed when one port was at a slower speed. This includes connecting plenty of fast ethernet clients on to the various gigabit switches above as well as even a few 10Mbps clients in a couple of the switches (VOIP gear).

Pretty much all of my experience is on small/cheap switches. The TP-Link SG2216 I mentioned above, which is my home networks current core switch, is my only experience with any kind of "fancy pants" managed/L2 switch (and its pretty nice). All the rest have been 5/8/16 port dumb switches. Zero experience with >16 port switches and none with any other managed switches.

 

[Nerre]

If I recall correct at least some cheaper 10/100 switches used a kind of cheating when it comes to speed.

Instead of having each port capable of 10/100 the switch had two "sides" internally, one 10 and one 100. Then each port was "assigned" to either side.

This meant that there was only ONE channel between the 10 side and the 100 side, so all devices using 10 Mbit/s would share and bandwidth of 10 Mbit/s towards the 100 Mbit/s side.

So, two 10 Mbit/s devices talking to two 100 Mbit/s devices (one each) would only get 5 Mbit/s throughput becuase of this shared channel.

A 10 Mbis/s device talking to another 10 Mbit/s device would get the full speed even if there was other traffic on the switch.

 

[Terry Kennedy]

If I recall correct at least some cheaper 10/100 switches used a kind of cheating when it comes to speed.

Instead of having each port capable of 10/100 the switch had two "sides" internally, one 10 and one 100. Then each port was "assigned" to either side.

This was popular quite some time ago, before switching chips with lots of ports were available. An example would be the Netgear DS1xx product line. But Netgear clearly marketed them as "Dual Speed Hub", not as a switch.

Unmanaged switches like the Netgear GS1xx family are pretty reliable. The only problem is that they can stop functioning and without monitoring software, you won't know where the problem is. Both the Netgear and Trendnet units I've used have some sort of problem where they stop forwarding when they see certain packets. I can reliably trigger it by simply connecting an unconfigured Cisco router or switch anwhere on my network. This makes me think it is a broadcast/multicast packet that's confusing them, and that they both use a switching chip from the same manufacturer. What is particularly annoying is that the switch appears to be acting normally - the port LEDs are blinking, but the switch isn't doing any packet forwarding.