Ask Your Question
0

One way throughput problem

asked 2024-09-21 19:30:31 +0000

quazi gravatar image

updated 2024-09-21 21:38:14 +0000

Chuckc gravatar image
Connectivity:
Host A ----- Switch ----- Router ----- Router ----- Switch ----- Host B
IP:
Host A: 10.10.10.1
Host B: 10.10.10.2

Host A and B are connected through Router and Switch over an EVPN VPWS tunnel. When transferring data from Host B to Host A, not getting throughput more than 20 Mbps in a single session. However, when transferring data in the opposite direction from Host A to Host B, getting 97 Mbps in a single session. There is no packet loss between Routers. Need help to troubleshoot one way throughput issue.

Capture for file transfer from Host B to Host A:
Host A: https://drive.google.com/file/d/1l1yr...
Host B: https://drive.google.com/file/d/1zBp-...

Capture for file transfer from Host A to Host B:
Host A: https://drive.google.com/file/d/1tlt3...
Host B: https://drive.google.com/file/d/1aBDv...

Thank You.

edit retag flag offensive close merge delete

Comments

Those capture files (B>-A files > 200MB; A->B files > 600 MB) are kinda big for analysis.
Can you describe the "transferring data" (protocol, application, type of data) to give some insight to how the files might be pared down.

"Upload_from_10.10.10.2-Capture_from_10.10.10.1.pcapng (241M) is too large for Google to scan for viruses. Would you still like to download this file?"

Chuckc gravatar imageChuckc ( 2024-09-21 21:43:57 +0000 )edit

Good day Chuckc.

File transferred over HTTP (TCP, Port 80) using HTTPFileServer.

For the file transfer from Host B > Host A, I noticed a very high number of tcp.analysis.flags on both sides, but at Host A it is higher (55%) than at Host B (27%). At Host A, TCP Out-of-Order is also relatively high.

However, tcp.analysis.flags for the file transfer session from Host A > Host B are 7% and 2%.

Suspecting Tx path issue from Host B > Host A. Need guidance to identify the actual point of failure and the root cause.

quazi gravatar imagequazi ( 2024-09-22 03:45:59 +0000 )edit

hi

If you plug in Host C in to the same switch where Host A is connected are you able to reproduce the problem? can you do the same test but this time connect Host C to the switch where Host B is connected.

Also can you use iperf for your thruput test and post the results?

net_tech gravatar imagenet_tech ( 2024-09-22 18:56:06 +0000 )edit

Upload_from_10.10.10.2-Capture_from_10.10.10.2.pcapng - 3 way handshake isn't captured

Looks like this is the HTTP server OP is using - > https://sourceforge.net/projects/hfs/...

net_tech gravatar imagenet_tech ( 2024-09-22 20:54:51 +0000 )edit

Hi @net_tech Trying to share the result of the suggested environment using Host C at both side and iperf.

Yes, HFS HTTTP file server used for throughput test.

quazi gravatar imagequazi ( 2024-09-23 08:44:30 +0000 )edit

What Windows OS is being used on a server where HFS is installed / running? Can you also run iperf between hosts A and B in both directions and share the capture results

net_tech gravatar imagenet_tech ( 2024-09-23 10:40:10 +0000 )edit

Also, when making new packet captures, please start your capture before doing the tests so that the 3-way handshake will be captured too.

SYN-bit gravatar imageSYN-bit ( 2024-09-23 13:28:54 +0000 )edit

@net_tech both side OS is 64bit Windows 10. Will share capture for iperf.

@SYN-bit, noted.

Shall I use iperf with the default parameters or any specific requirement?

quazi gravatar imagequazi ( 2024-09-24 07:57:21 +0000 )edit

yes, defaults are fine. -s for server -c for client

net_tech gravatar imagenet_tech ( 2024-09-25 03:18:14 +0000 )edit

@net_tech and @SYN-bit The pcap files for iperf TCP and UDP 10 seconds session between Host A <> Host B stored as compressed file (.rar) in the following google drive url.

https://drive.google.com/file/d/1W9e6...

@net_tech The pcap files for iperf TCP and UDP 10 seconds session between Host A <> Host C connected in the same switch are uploaded as compressed (.rar) in the following google drive url. As Host B is located in the remote site, not able to generate Host B <> Host C on the same switch scenario.

https://drive.google.com/file/d/1ssN0...

Before uploading in google drive, both the files were checked with antivirus and found clean.

OS:- Host A: 64-bit Windows 10, Host B: 64-bit Windows 11, Host C: 64-bit Windows 10

Thanks for your time.

quazi gravatar imagequazi ( 2024-09-25 17:36:01 +0000 )edit

You are getting almost wirespeed when you are testing A->C (984Mbps)

A Hardware: 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz (with SSE4.2) OS:64-bit Windows 10 (2009), build 22631 Application: Dumpcap (Wireshark) 3.0.6 (v3.0.6-0-g908c8e357d0f)

C->A Is slightly less, negligible difference on the grand scheme of things (905Mpbs). Here you have 0.1% of TCP Window Full (260 out of 202,263 packets) where host C is telling host A to hold up as it can't empty the buffer fast enough to receive more data.

C Hardware: Intel(R) Core(TM) i5-1035G1 CPU @ 1.00GHz (with SSE4.2) OS:64-bit Windows 10 (22H2), build 19045 Application: Dumpcap (Wireshark) 4.0.2 (v4.0.2-0-g415456d13370)

If you can do the same test at the remote site and get B<->C iperf tests, we would be able to rule out host B. Until then the ...(more)

net_tech gravatar imagenet_tech ( 2024-09-28 22:41:02 +0000 )edit

1 Answer

Sort by ยป oldest newest most voted
0

answered 2024-09-23 13:37:35 +0000

SYN-bit gravatar image

From a quick glance on the pcap files, I suspect the following:

  • There is quite a bit of packet reordering of the data packets from 10.10.10.2 (Host B) to 10.10.10.1 (Host A).
  • The reordered result in DUP-ACKs
  • Due to the amount of reordering, there are tripple DUP-ACKs, resulting in fast-retranmissions
  • The congestion window will be kept low when there are retransmissions, assuming there was packet-loss due to congestion

Are you able to capture on the WAN (EVPN) side of the routers?

edit flag offensive delete link more

Comments

Not able to capture the WAN side of the routers due to high volume of traffic, 70Gbps usages on 100Gbps link. Only having problem for a specific LAN side connectivity.

quazi gravatar imagequazi ( 2024-09-24 09:28:43 +0000 )edit

you don't need to capture the payload, limit your capture buffer to be slightly larger than the packet header.

net_tech gravatar imagenet_tech ( 2024-09-28 22:46:11 +0000 )edit

Noted with thanks @net_tech.

quazi gravatar imagequazi ( 2024-09-29 12:01:17 +0000 )edit

is MTU set to 9000 on both A and B hosts?

net_tech gravatar imagenet_tech ( 2024-09-29 15:28:21 +0000 )edit

No, both the hosts are in windows default configuration. Host A connected switch jumbo enabled but Host B connected switch in default MTU 1500 and both the routers upstream and downstream ports configured with 9216 MTU.

quazi gravatar imagequazi ( 2024-09-29 16:22:46 +0000 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

1 follower

Stats

Asked: 2024-09-21 19:30:31 +0000

Seen: 348 times

Last updated: Sep 23 '24