Network switches reducing throughput

[sxr7171]

I understand theoretically and also in most cases practically that connecting a network switch shouldn’t reduce throughput.

I have a Netgear R7800 connected to a 400mbps service which delivers 450mbps peaks.

I have tried two switches:


Dlink GO-SW-5G

It is cheap and possibly suspect so after checking cables.

I bought:

Netgear GS308

Also cheap but gets good reviews and looks built well.

They both cut speeds down to the PC attached to it. I get better speeds on Wireless AC.

The cut isn’t a whole lot but it’s maybe 5-10% which is really high for something that should be full duplex 1gbps.

Is this normal? Are the switches I buy too cheap?

For now I can connect directly to the router but for future knowledge I’d like to get to the bottom of this.

 

[VirtualLarry]

That sounds kind of "internet speedtest margin-of-error" kind of thing.

Can you set up two host PCs, connected via 1Gbit/sec LAN connections, through each of those switches, and run IPERF?

Or set up a RAMdrive or SSD, and run some LAN file-copy tests?

Internet speedtests can be less than conclusive.

 

[JackMDS]

I

The cut isn’t a whole lot but it’s maybe 5-10% which is really high for something that should be full duplex 1gbps..

If it was me I would feel so lucky that I would Run and buy "'Power Ball tickets".

The published "Marketing" rating of Network related appliances is the rating of the of the core chipset of the the appliances. It is Not the actual overall performance in and out of the whole "Box".

Actual performance always lower than the pure rating of the Chipset, in many cases it is more than 10%.

 

[VirtualLarry]

If it was me I would feel so lucky that I would Run and buy "'Power Ball tickets".

The published "Marketing" rating of Network related appliances is the rating of the of the core chipset of the the appliances. It is Not the actual overall performance in and out of the whole "Box".

Actual performance always lower than the pure rating of the Chipset, in many cases it is more than 10%.

You're right, Jack, but it's been my experience so far, buying "cheap" $20 Gigabit 8-port switches (mostly from TrendNet, some Netgear), that consumer 1Gbit/sec switches have practically negligible overhead.

Meaning, if you have Intel NICs on either side, that can push file transfers of 114MB/sec, then you'll still get that through a consumer switch, or even 3-4-5 of them. At least, for me, that's been my experience.

Now, SOHO routers, that's another thing entirely, as some of them, intercept every packet coming and going on WAN and LAN ports (often, part of the same ethernet switch, just segregated via VLANs on the switch unit), and the router's CPU processes it, or can. Most don't, though, and can sustain line-rate LAN transfers via their LAN ports between PCs with 1GbE NICs.

I still think that this is down to "internet variance", and OP should do some controlled switch testing with multiple PC hosts with 1GbE NICs, and IPERF or netbench or something like that.

 

[mxnerd]

Connection speed <> throughput. I wouldn't believe that wifi performs better than wired ethernet if the client has only one antenna.

Toms Hardware tested 4 switches and Netgear GS308 performed best.
Direct wire vs swicth was 928Mbps/944Mbps = 98.3%

https://www.tomshardware.com/reviews/unmanaged-gigabit-ethernet-switch-roundup,4321-5.html

Test local speed using internet website like http://speedtest.net also results incorrect info.

 

[dave_the_nerd]

Try a different cable?

 

[sxr7171]

Cables were checked. But we are not talking about gigabit speeds we are just talking about 450mbps being cut down to maybe 390-420mbps.

 

[mv2devnull]

Just to be clear, you have compared?
A) PC --cable-- Router
B) PC --cable-- Switch --cable-- Router

 

[VirtualLarry]

I have a Netgear R7800 connected to a 400mbps service which delivers 450mbps peaks.

Cables were checked. But we are not talking about gigabit speeds we are just talking about 450mbps being cut down to maybe 390-420mbps.

I think that Comcast still does this, they give you a "Boost" for initial downloads and speedtests.

If your service is 400Mbit/sec, and you're getting 420Mbit/sec at times, then what is the problem?

Again, don't use internet speedtests that can be variable, to test LAN speeds. Use a controlled test like IPERF.

 

[ch33zw1z]

I think that Comcast still does this, they give you a "Boost" for initial downloads and speedtests.

If your service is 400Mbit/sec, and you're getting 420Mbit/sec at times, then what is the problem?

Again, don't use internet speedtests that can be variable, to test LAN speeds. Use a controlled test like IPERF.

Yes, this is also my suggestion. Use iperf to test LAN to LAN communications. You can swap out switches, cables, etc....to see which provides the best performance.

 

[mnewsham]

Trying to diagnose local network equipment with WAN tests is silly, iperf as others have suggested.

 

[sxr7171]

Ok thank you. I’ll get IPERF.

 

[sxr7171]

Just to be clear, you have compared?
A) PC --cable-- Router
B) PC --cable-- Switch --cable-- Router

Yes.

 

[mxnerd]

Like all others have said, testing speed using WAN is meaningless. You can't control anything beyond your router. You have to test the speed locally with another PC (using IPERF or other tools).

 

[Rifter]

You're right, Jack, but it's been my experience so far, buying "cheap" $20 Gigabit 8-port switches (mostly from TrendNet, some Netgear), that consumer 1Gbit/sec switches have practically negligible overhead.

Meaning, if you have Intel NICs on either side, that can push file transfers of 114MB/sec, then you'll still get that through a consumer switch, or even 3-4-5 of them. At least, for me, that's been my experience.

I agree, i happen to own a few older GS308's and they both hit 110-114MB/s peak transfers over my network.