[SOLVED] Strange upload performance with Gigabit FTTH connection

bmeeks

I am experiencing a strange upload speed test problem with my relatively new Dell PC and FTTH symmetrical Gigabit service. This will take a bit to fully explain. I'm at my wits end trying to diagnose what might be wrong.

I have a Netgate SG-5100 for my home network firewall. I have three physical PCs in the home and some virtual machines hosted in an VMware EXSi hypervisor. Everything is hard-wired using Cat5e cabling to a Netgear 24-port Gigabit switch (the LAN side).

The SG-5100 LAN port connects to the Netgear switch. The SG-5100 WAN port connects to the ISP's ONU. My Internet service is symmetrical Gigabit fiber-to-the-home with the 1000/1000 plan. A Cat5e cable runs from the ISP ONU (an Azores Networks WAG-20D) Gigabit LAN port directly into the SG-5100's WAN port (an Intel igb port).

The three physical PCs on my LAN are a Dell XPS 8940 (purchased last year), a Dell XPS 8950 (just purchased and installed last month), and an older HP desktop (circa 2018). The two Dell PCs run Windows 11, and the HP desktop PC runs Windows 10.

Speed tests run from the newest Dell XPS 8950 show 950+ megabits/sec down and up to the ISP's internal test server. Ditto for the older HP deskop. And also see similar speeds when testing from my ESXi virtualized machines. All get 950+ megabits/sec on a fairly consistent basis to the ISP's internal Open Speed Test server. However, my Dell XPS 8940 always gets 950+ megabits/sec down, but never exceeds 200 megabits/sec up! And yes, I've swapped out the cable, physically moved the PC and plugged it into the port that the newer XPS 8950 was plugged into (to eliminate the switch port as a cause), and nothing changed.

But here is the really puzzling point. I installed the same speed test software my ISP is using (Open Speed Test server) on the older HP desktop so that now I have a speed test server on my internal LAN that is identical to the one the ISP has on their network. All of my machines, including my "problem XPS 8940", get consistent 970+ megatbits/sec UP and DOWN test results to my internal LAN server. BUT that "problem XPS 8940" never gets more than 200 megabits/sec up (and usually closer to 160 megabits/sec) when testing outside my LAN.

So, what is going on here? How can the Dell XPS 8940 easily get 980 megabits/sec both up and down when testing to my local LAN-based Open Speed Test server, but then get 970 down from my ISP's test server but only 160 up? I don't think it is congestion on my ISP's network because immediately after getting a poor upload with my XPS 8940 PC, I can test using the XPS 8950 PC (or even the older HP desktop) and get 950+ both up and down to the ISP's server.

If I had a bad NIC in the XPS 8940 machine, or a bad switch port, or a bad cable, then I would expect the 8940 machine to show poor speed test results to both my internal LAN speed test server and to the ISP's speed test server. But that's not what I see. I get excellent speeds when testing to my local LAN speed test server and also when using iperf3 testing between the Dell XPS 8940 and any other machine on my LAN. It's only when testing to external speed test sites that I get poor uploads from the XPS 8940 machine.

Does anybody have a suggestion for where to look for the problem? The only thing materially different with the XPS 8940 is it has an installation of VMware Workstation 17. None of the other machines have that. But if VMware Workstation was the problem, I would expect it to adversely impact the speed tests to internal servers on my LAN. But I am not seeing that.

johnpoz

@bmeeks Dell 8940... I have the same machine ;)

I don't have that sort of internet connection to test with.. But I see no problems locally.. Get 2.37 gig via a cheap usb adapter even both up and down via iperf.. And gig locally is fine as well. same 950 up/down

- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.00  sec  1.11 GBytes   952 Mbits/sec                  sender
[  5]   0.00-10.02  sec  1.11 GBytes   949 Mbits/sec                  receiver

That is the 8940 sending to the nas.

My 8940 came with some killer nic software for the killer e2600 nic it came with - I have that all turned off and uninstalled.. Does your still have that bloatware installed?

What do you have hooked into the nic, you mention vmware software... I would prob test removing any sort of bridge software like vmware or other drivers..

example - while am not seeing any issues with either of those hooks, I have seen hooks cause problems in the past.. Locally can sometimes be way different in that the latency is so much faster, etc. and some sort of issues or mtu sort of things might not show up.. Can't hurt to remove any such things from your nic for testing.

bmeeks

@johnpoz said in Strange upload performance with Gigabit FTTH connection:

@bmeeks Dell 8940... I have the same machine ;)

I don't have that sort of internet connection to test with.. But I see no problems locally.. Get 2.37 gig via a cheap usb adapter even both up and down via iperf.. And gig locally is fine as well. same 950 up/down
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.00  sec  1.11 GBytes   952 Mbits/sec                  sender
[  5]   0.00-10.02  sec  1.11 GBytes   949 Mbits/sec                  receiver
That is the 8940 sending to the nas.

My 8940 came with some killer nic software for the killer e2600 nic it came with - I have that all turned off and uninstalled.. Does your still have that bloatware installed?

Thanks for replying! Yes, I have tried both with and without the Killer Intelligence Center bloatware installed. No difference. My internal iperf3 tests mirror yours above. It's blowing my mind that the machine works fine internally, but has issues going external when no other machine on my LAN has issues external.

As expected, different test times to the ISP's internal test server do yield slightly differing speed results from my internal machines. But overall, all my machines EXCEPT the XPS 8940 do well externally. The 8940 is my daily-use machine in my home office, so that's why it bothers me for it to fail so miserably .

bmeeks

@johnpoz:
Replying to your follow-up about bridge software.

That's my next test, but I do use the VMware Workstation VMs a lot. That's where I have a pair of pfSense machines that I do all of my IDS/IPS package development testing on. I could move them to the ESXi host I guess.

I am thinking the latency is a key here. Obviously my local network has super low latency. I'm thinking something is triggering the Windows congestion algorithm to misbehave. I did a Wireshark capture and did see a ton of TCP Retransmissions and Dup Acks on the 8940.

Guess just to prove to myself where the issue is, I will remove VMware Workstation and test.

johnpoz

@bmeeks I take it you don't have any sort of qos or shaping enabled on your pfsense.. your not running something like cFosSpeed on the box are you - have seen that be problematic for users forget they set it ;)

I take it you did a reset with netsh?

I really wish I had a gig up to be able to try and duplicate your issue, but I am stuck with only 50 up that have no problems with, I do see 60 up now and then ;)

And download always exceeds my isp 500, normally see in the 550ish range.. But your issue is on upload.. if download I would think maybe recv window..

I take it as well you have tried other test sites - fast.com for example?

johnpoz

@bmeeks said in Strange upload performance with Gigabit FTTH connection:

will remove VMware Workstation and test.

you shouldn't have to remove it - just unbind whatever its bridge stuff is from your nic.

johnpoz

@bmeeks hmmm - just for grins you want to compare settings

$ netsh int tcp show global                         
Querying active state...                            
                                                    
TCP Global Parameters                               
----------------------------------------------      
Receive-Side Scaling State          : enabled       
Receive Window Auto-Tuning Level    : normal        
Add-On Congestion Control Provider  : default       
ECN Capability                      : enabled       
RFC 1323 Timestamps                 : disabled      
Initial RTO                         : 1000          
Receive Segment Coalescing State    : enabled       
Non Sack Rtt Resiliency             : disabled      
Max SYN Retransmissions             : 4             
Fast Open                           : enabled       
Fast Open Fallback                  : enabled       
HyStart                             : enabled       
Proportional Rate Reduction         : enabled       
Pacing Profile                      : off

bmeeks

Here is the netsh output:

PS C:\Users\bmeeks> netsh int tcp show global
Querying active state...

TCP Global Parameters
----------------------------------------------
Receive-Side Scaling State          : enabled
Receive Window Auto-Tuning Level    : normal
Add-On Congestion Control Provider  : default
ECN Capability                      : disabled
RFC 1323 Timestamps                 : allowed
Initial RTO                         : 1000
Receive Segment Coalescing State    : enabled
Non Sack Rtt Resiliency             : disabled
Max SYN Retransmissions             : 4
Fast Open                           : enabled
Fast Open Fallback                  : enabled
HyStart                             : enabled
Proportional Rate Reduction         : enabled
Pacing Profile                      : off

These are the defaults (or at least I don't recall changing any). Do see a couple of differences from yours, though. Will try making some changes to see the impact.

johnpoz

@bmeeks generally timestamps are a good thing, but If I recall why I changed that is locally was seeing performance hit..

bmeeks

Disabling the VMware Bridge setting in the NIC connection properties seemed to help for one test. I got several hundred megabits/sec UP (like 600ish). That's way above the 150 -160 I was typically seeing.

But I could not repeat that result. Subsequent tests fell back to the 180 -190 range at best. VMware Workstation installs several virtual NICs as well. So all of that may be the culprit. Not too big of a deal to remove, test, and then reinstall.

I did try the network stack reset command, but it made no difference. All it changed was altering the ECN Capability back to "disabled" from the "enabled" I had tried based on the delta between your settings and mine in our initial comparison.

I'm not enough of a TCP guru to explain why, but my gut thinks the small increase in latency with the external server is playing a role here. Will take me a few minutes to remove VMware Workstation and test that way. I will post back with the results. This has the potential to become an obsession , but it's driving me crazy to get such conflicting results!

johnpoz

@bmeeks said in Strange upload performance with Gigabit FTTH connection:

This has the potential to become an obsession

I know exactly how you feel - I just wish I had the internet connection to try and help you ;) But I don't have any way to try and duplicate your problem.. For all I know would be seeing the same thing if I had upload of gig, and just not seeing the issue because I only have 50 up.

But if your seeing a load of retrans and such on your sniff - for you upload, that isn't going to bode will for great speed that is for sure.

Only thing I can say is I sure am not seeing any upload issues locally with my 8940, and the killer e2600 nic, I am running windows 10 22H2, build 19045.2604

johnpoz

@bmeeks you maybe have already done this - but when I suspect its something with windows or the driver for windows. To rule out an actual hardware or networking issue.. I like to boot a linux distro.. This way no changes to anything to do with your windows box, but you rule out anything in the path, and pretty much pinpoint something to do with windows or the windows drivers..

bmeeks

Yeah, booting to a Linux distro is a good idea. Will make a disc and try that out.

I've removed VMware Workstation, went through and cleaned just about everything related to Dell bloatware including all the Killer Performance stuff. Still no improvment for external upload speeds, but testing to my internal box is still fine.

Let me make a Linux Live CD and test that. This has become a quest for sure.

I was totally surprised to have Gigabit fiber suddenly available to me here in my little rural town. Turns out there will soon be two competing providers in my neighborhood. The federal and state governments are handing out "rural broadband grants", and they are having an impact. The local electric EMC co-op is stringing fiber over their entire four-county coverage area. They started in their home county, which is where I live. But for my neighborhood, a local provider guy I know got the jump on the EMC. They started working in my neighborhood in January this year doing everything underground. I got my service connected last Tuesday afternoon.

They have done a ton of directional boring with one of those fancy tunnel machines. Of course, they did manage to puncture the city's water main in the neighborhood 3 different times in different places over the four weeks they were drilling. That may have eaten up the first year's anticipated profits for sure .

johnpoz

@bmeeks said in Strange upload performance with Gigabit FTTH connection:

I was totally surprised to have Gigabit fiber suddenly available to me

I keep checking - so the big question, do you have a public IPv4 or you behind a cgnat?

I could easy go gig down with my isp, or a couple of others - but I don't really care about down, 500 is fine for my needs.. But I really want more up.. 50 up is a bottle neck to my plex.. While I have not really run into any real issues as of yet.. I just wish is was 500/500

bmeeks

@johnpoz said in Strange upload performance with Gigabit FTTH connection:

@bmeeks said in Strange upload performance with Gigabit FTTH connection:

I was totally surprised to have Gigabit fiber suddenly available to me

I keep checking - so the big question, do you have a public IPv4 or you behind a cgnat?

I could easy go gig down with my isp, or a couple of others - but I don't really care about down, 500 is fine for my needs.. But I really want more up.. 50 up is a bottle neck to my plex.. While I have not really run into any real issues as of yet.. I just wish is was 500/500

CGNAT for now, but their engineer promised they were working on direct public IPs. He did offer to configure a static port forward if I wanted, but I deferred. They are also getting ready to roll out an IPv6 BETA, and he will enroll me in that. This is a small town ISP, and I personally know the folks, so that is a positive that helps override the negative of CGNAT.

Like you, I wanted more upload. I was also limited to 50 meg with my cable ISP. A symmetrical path is nice -- no more bufferbloat shaping needed (well, at least for most of my machines) - this problem with my 8940 notwithstanding. Even still, with 150 - 180 megabits of upload, I score an A on the bufferbloat test. Haven't tested from one of my machines that gets full bore upload. Should get A+ there.

johnpoz

@bmeeks they could give me 10ge up and it wouldn't do me much good for plex ;) heheh if behind a cgnat.

bmeeks

@johnpoz said in Strange upload performance with Gigabit FTTH connection:

@bmeeks they could give me 10ge up and it wouldn't do me much good for plex ;) heheh if behind a cgnat.

I hear you. The only thing I did remotely was inbound VPN for rare remote access. But with secure cloud storage a bit more mature, I can now put the MS Word docs my wife and I might want to access remotely in the cloud. That cut down on the need for inbound VPN.

I get that it's hard for these small ISPs to afford enough IPv4 space to provide public routable IP addresses for all their customers.

johnpoz

@bmeeks well with that gig down you should have like 20 or 30 linux distros download and ready - have you tested it yet?

bmeeks

Working on it. Waiting on the bootable USB utility to finish creating a USB image. Decided to go that route instead of a DVD.

It's just finishing verifying, so going offline and going to boot into Ubuntu and test. Will report back when I'm done.

SteveITS

@bmeeks
I'd start with the "Receive-Side Scaling" and/or "Receive Segment Coalescing" settings. I seem to recall occasional posts in the past. Couldn't find any here quickly, but found:

This is a Dell "server" document: https://www.dell.com/support/kbdoc/en-us/000138508/how-to-resolve-performance-tuning-and-connectivity-issues

random person:
https://forums.cox.com/forum_home/internet_forum/f/internet-forum/24208/to-all-the-people-that-have-been-experiencing-low-upload-speeds-i-have-had-a-break-through