Local Network Transfer Killing Internet Speed

Soldato
Joined
21 Jan 2010
Posts
24,516
Hi folks,

Really silly question (given I am an [expired] CCNP) :D.

I have spent today rewiring my 'comms cupboard'. My main PC connects to Switch A. My server connects to Switch B. My router connects to Switch A. I am transferring a bunch of binaries to the server from my main PC. The transfer rate is approx. 10MB/sec (limited by HDD).

For some reason, and for the first time in my long-term memory, I am seeing extremely slow internet whilst transferring. Is Win 11 doing something odd? Is it because I am across switches? Ping is through the roof too.

This is a cheapo TP Link Gigabit switch. Perhaps I am saturating the switch?

Edit: I just realised Switch B (my PoE switch) is only 10/100. I've put them both on the same switch now. It is odd because my router connects to Switch A directly; so I wasn't expecting speed issues.

Edit2: I've put both PC and Server to Switch A. Ping is at 600ms just getting to the internet router. I know my switch is garbage but is it really that bad? (TP Link LS1005G).
 
Last edited:
Your transfer speed is consistent with a 100 Mb link. Go into the properties of the NIC on both your server and workstation and explicitly set them to gigabit.
 
Your transfer speed is consistent with a 100 Mb link. Go into the properties of the NIC on both your server and workstation and explicitly set them to gigabit.
My transfer speed was inline with what I expected for a standard HDD - in all fairness, you've made me realise the server is 10/100 (a job for later). However, that shouldn't mean my ping goes up to 400ms+ and my internet slows down from my main desktop (10/100/1000). Must be the cheapo switch?
 
You need to check that your desktop is operating at full gigabit speeds. If it's working at 100 Mb/s then you are saturating the link and your ping will indeed go to pot.

The speed you are getting is not consistent with a HDD: you should be getting 100 MB/s or more, 8-10 times what you are getting.
 
You need to check that your desktop is operating at full gigabit speeds. If it's working at 100 Mb/s then you are saturating the link and your ping will indeed go to pot.

The speed you are getting is not consistent with a HDD: you should be getting 100 MB/s or more, 8-10 times what you are getting.
The desktop is full duplex gigE and operating as such. The server is 10/100. So I shouldn't be saturating the desktop at all (but yes the server should be getting pwnt).

Is it really that high nowadays? I remember having to build a ZFS fileserver to get anywhere close lol! So out of touch.

Did you dump your CCNP? :cry:
I know right :cry:. In my defense I used to do TSHOOT to renew so I haven't touched switching for a good 15 years :cry:

It's gotta be the switch right? I have no other explanation other than I'm saturating the switch!!!
 
I know right :cry:. In my defense I used to do TSHOOT to renew so I haven't touched switching for a good 15 years :cry:
TShoot is the easiest way to renew from what I've been told (my old housemate is a CCIE), although I admit I never sat any CCNP level certs. I did ICND1 and then got an email to say it's expiring so I sat ICND2 without revision and passed it. I'd been doing CCNP level stuff for almost 15 years and never had an issue getting work so figured what's the point.
 
It's a 5 port gigabit switch, they are much of muchness tbh, you could try cabling directly into your router/hub and see what happens. Have you replaced all cables as well?
Yeah this was my assumption as well. It was only 9 quid though.... I might upgrade to the 8 port as I'm short anyway, and see how I get on. It's so painless to boomerang it back to Amazon I'll just start there.

TShoot is the easiest way to renew from what I've been told (my old housemate is a CCIE), although I admit I never sat any CCNP level certs. I did ICND1 and then got an email to say it's expiring so I sat ICND2 without revision and passed it. I'd been doing CCNP level stuff for almost 15 years and never had an issue getting work so figured what's the point.
But how will you know all of the Cisco proprietary standards? :cry:. I did it in my spare time hot on the back of CCNA and never actually got a job in that field at all.
 
But how will you know all of the Cisco proprietary standards? :cry:. I did it in my spare time hot on the back of CCNA and never actually got a job in that field at all.
Standards actually came up in work last week, even though I am now a server/cloud nerd I still get involved with networks when people don't know what they're doing. Not mentioning any company names but 'they' forgot that Cisco's RPVST+ and industry standard RSTP don't talk to one another unless you tell the switches to do so, created a loop which took down a hospitals phone system including their 999 call routing. :cry: I rocked up, 'did you tag VLAN 1 on the Cisco to 'other vendor' switch port-channels?'. I guess not...
 
I am also wondering @ChrisD. whether my terminations may be an issue here. I have crimped 10's of thousands of RJs, but again - not for over a decade. I also never had to use CAT6 and in the instance, it is CAT6. To get a tight crimp with the cable holder distinctly on the casing, it is a fraction trickier. Plus I am under the stairs and in a confined space.

It is possible a bad crimp is at play? Is there a way to test between two end points?

Though, I do have two runs to the computer so I can try left or right; and I am pretty sure I had the same behaviour....

Edit: I get <1ms ping to the router, and 600ms when transferring.
 
Last edited:
Problem "solved" - I swapped out the cheapo (£9, lol) TP Link switch for a Zyxel. I'm transferring 2gb at 100MB/s and the latency to the router has only kicked up a fraction (1ms -> 60ms).
 
For some reason, and for the first time in my long-term memory, I am seeing extremely slow internet whilst transferring. Is Win 11 doing something odd? Is it because I am across switches? Ping is through the roof too.

I *think* you're experiencing something called "head of line blocking" which is something you normally find on older cheaper switches with poor congestion management.

Essentially what I think is happening, is that the 100Mb interface connected to your server is probably getting close to 100% average usage. Because your input interface from your PC is 1Gbps, there will be moments where you're either on the 100Mb limit, or oversubscribing the slower server interface momentarily.

When this happens, the switch will send a "backpressure signal" towards the ingress port (the 1Gbp port from your PC) asking it to slow down by buffering and dropping all packets arriving ingress, from your PC. This will bugger up everything, even if its only a couple of times per second, as you'll end up with your TCP acks getting dropped, things going out of sequence, and retransmissions will create more pain.

This means, that even if the 1Gb ports between your PC and router have 900Mbps spare bandwidth in theory, in reality - they don't because any congestion occurring on the slower server port, will force the switch to slow the whole system down, which includes randomly dropping packets from all ingress flows on that 1Gb port, not just ones destined to the server.

This type of problem is mostly solved in newer hardware, where the switch ports drop on egress port congestion, or some other method (store/forward) - or in more expensive kit, virtual-output queues which have a bit more intelligence.

(apologies for the nerd answer :p )
 
TP Link LS1005G. It suggests a 10gb backplane according to some chap on Reddit but clearly couldn't handle transfer and internet use together.

Backplane is never a problem on these things - it's a standard non-blocking switch, you could plug a PC into each of the 5x ports at 1Gbps and get line-rate on all of them, and not have any drops.

The problems start, where you have some ports running at 1G and some running at 100Mb (speed step), as it creates a situation inside the device, where a faster port is sending traffic to a slower port - congestion. If the device is only £9 and doesn't have any proper congestion control (which it won't for that price) it'll just buffer the ingress port to hell, until the whole system slows right down.

I'd bet, that if you changed your server port to 1G (so all ports on the switch were 1G) then the problem wouldn't be there.
 
Backplane is never a problem on these things - it's a standard non-blocking switch, you could plug a PC into each of the 5x ports at 1Gbps and get line-rate on all of them, and not have any drops.

The problems start, where you have some ports running at 1G and some running at 100Mb (speed step), as it creates a situation inside the device, where a faster port is sending traffic to a slower port - congestion. If the device is only £9 and doesn't have any proper congestion control (which it won't for that price) it'll just buffer the ingress port to hell, until the whole system slows right down.

I'd bet, that if you changed your server port to 1G (so all ports on the switch were 1G) then the problem wouldn't be there.
Makes sense! Thank you for taking the time to explain. Hadn't really considered that at all. I'm pained to buy a 1gbit adapter for it as checking my Amazon history I should have at least 5 knocking around, lol.
 
Back
Top Bottom