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

Slow application response times

1
1

I'm troubleshooting a slow performing application which is developed by a third party.

Info:

  • Client/server application setup
  • End-users are located in Site A with client app installed on their physical Pc's
  • Server is located in Site B, accessible via a private 1Gbps circuit
  • Circuit utilization is low/normal
  • No other issues reported

  • The response times are circa 10000ms

  • Slow response times only noticed on physical desktop Pc's
  • Issue not present on thin client sessions (thin client server is located in Site B)

q

q

q

I am looking for assistance in ruling in (or out) the network and any other pointers if visible from these traces

Thank you

asked 13 May '16, 08:27

Brad_101's gravatar image

Brad_101
415811
accept rate: 0%

edited 16 May '16, 03:41

Do you mean with response time, the time the test need to run? On a physical client it takes 13s and on a Thin Client the test needs 6s to run?

BTW: My personal understanding of response time, is the time the application needs to response to the request.

(13 May '16, 12:25) Christian_R

That is correct. The time for the test to run and it seems to be physical client only

(13 May '16, 12:36) Brad_101

Well the thin clients are connected server. So if I understand it right the terminal server is located in the same location (Datacenter) like the Oracle?! So only the graphical output must be transfered over WAN, or am I wrong?

(13 May '16, 12:47) Christian_R

That is correct however I have been advised the same test has been conducted from a physical pc in another site (all three in a triangle) and the response time was good.

I haven't seen that for myself Though

(13 May '16, 12:59) Brad_101

In the trace we see for the terminal server <1ms network delay for the physical client we see up to 8ms delay between request and response at the client side.

(13 May '16, 12:59) Christian_R

2 Answers:

1

Well in the trace of the terminal server we can see test duration of 1.3s (Paket 1225 to 4399) and in the physical we can see a duration 11,2s (Paket 11250 to 14433) which is around 6 times more. We know that the Response time varies a little bit around 6 - 8 ms and of the terminal server, too. It could explain it from my point of view.

alt text

answered 13 May '16, 14:23

Christian_R's gravatar image

Christian_R
1.8k2625
accept rate: 16%

So, do you mean the delay is due to WAN?

(13 May '16, 14:29) Brad_101

Yes and 6-8ms round trip time is not so long for WAN from my point of view. The test scenario does not like this, by design.

(13 May '16, 14:31) Christian_R

Probably because the database query is badly written and requires say 1,250 request-response pairs to complete. 1,250 times 8 ms is 10s. On the LAN 1,250 times 0.8s is 1 second.

(13 May '16, 15:11) PaulOfford

It is exactly like Paul said. The database query is written so badly.

(14 May '16, 12:21) Christian_R

Why are the start & end times the same on client and server?

Wouldn't we see a longer period of time at the client?

If

(16 May '16, 09:17) Brad_101

@Brad_101

Your "answer" has been converted to a comment as that's how this site works. Please read the FAQ for more information.

(16 May '16, 09:36) grahamb ♦

No. You will see more or less the same graph on both ends, because the client first sends a new request after he has received the response of the old request.

(16 May '16, 09:47) Christian_R
(16 May '16, 10:15) Brad_101

No. There is a gap of more than 30 sec in the trace, where no data flows. I guess it is application or user related. But can´t say more, because the app data is sanitized. I talk about frame 956 and 057 of the server trace for example.

(16 May '16, 11:59) Christian_R

I see that now, thanks Christian.

That isn't user delay as I used that application during the capture & all that was done was run Wireshark, open app, wait for traffic to stop, open customer record, stop capture.

With TCP streams, how do you tell / know if in my example of opening a customer record, that conversation is or isn't a single TCP conversation?

(16 May '16, 12:24) Brad_101

I just guessed. But this looks like the one. Because it goes over 50s and looks a liitle bit like a download sessioon. But of course I am not sure, because the application data is missing. For 100% to be sure you always need the full socket pair.

(16 May '16, 12:56) Christian_R

Hopefully this is a better trace for you:

https://drive.google.com/open?id=0BxWWq7ozi-PFek1lbzBFcFVYS0U

(16 May '16, 13:25) Brad_101

Well this trace shows something new and something old. So could you please describe again in prosa the problem which we should see in the trace.

Because I don´t want explain things that does not matter in this case, like the zero windows.

(17 May '16, 11:56) Christian_R

This capture has approx a approx 35sec user delay and then a record was searched which took approx 50 sec to load. The 50 delay is the problem

(18 May '16, 04:50) Brad_101

Well as we cann see for StreamID==9 the RoundtripTime Grows during the session. Beware of, that we are now on the server side. The reason, if it is the network or the client can´t be reliable answered with this trace. But at least we can say it is not the server. alt text alt text

(18 May '16, 11:53) Christian_R
showing 5 of 15 show 10 more comments

0

It's probably unfaiir to say it's badly written without seeing the code. However, it's fair to say that it's written in a way that is not optimized for use over a WAN.

answered 16 May '16, 14:42

PaulOfford's gravatar image

PaulOfford
131283237
accept rate: 11%