This is a static archive of our old Q&A Site. Please post any new questions and answers at ask.wireshark.org.

TCP Window shrinking; Can Wireshark quantify the impact?

0

Hi y'all...

I am analyzing why a new data replication process is taking longer – and generating lower throughput – than is expected / acceptable.

I have looked at some packets, and I see the TCP Window on the “receiver” shrinking over time, and at times it shrinks effectively to zero, and the sender must wait until a TCP Window Update arrives, to resume sending. This occurs periodically.

So I know that the receiver is not able to “keep up” with the replication data being sent, which slows down the replication process.

What I do NOT know is: BY HOW MUCH??? If we are able to resolve this, would it improve throughput by 10%? 30%? 80%?

Does anyone have any suggestions on how I can use Wireshark's analysis capabilities to quantify the impact of this condition?

Thx all,

feenyman99

asked 28 Feb '17, 13:30

feenyman99's gravatar image

feenyman99
96222226
accept rate: 25%

Faster hard disk, more CPU, reduce system load that is what you can try.

(28 Feb '17, 13:47) Christian_R

The folks looking at the health of the receiving server tell me that CPU, Memory and Disk Utilization are all quite low during this data replication process. I am having someone take a second look to confirm this, but, as of now, there is no indication that the receiver is lacking capacity for this process.

Regarding sharing a trace... I will look at the link you provided (file sharing tutorial), as I'd like to know how to do this, but I must say that I'm in a VERY security-conscious industry sector, and I'm not sure I could share a trace file, even if I use a tool to modify/obscure the IP Addresses.

Thx!

(03 Mar '17, 09:15) feenyman99

One Answer:

2

What you can do to get the theoretical throughput is to calculate the Bandwidth-delay product (BDP; see. https://www.switch.ch/de/network/tools/tcp_throughput/index.html ).

Take the max reported window size reported by the sender and the initial Round Trip Time as input values.

With this you can calculate the diff to your measured throughput.

Please be aware that other influences (e.g. packet loss) can change the throughput too.

answered 01 Mar '17, 00:06

Uli's gravatar image

Uli
9031515
accept rate: 29%

edited 01 Mar '17, 00:07