Hi All.
I have a verrrrrry serious problem.
My company product is a server which sits between the Web-server and the internet. It is a kind of proxy.
We installed it at on of our customer which has a Solaris 2.6 Web-server.
Our Server is NT.
Immediately after installing our server, browsing of big files (> 30K) became very slow.
Below is a detailed description of the problem.
We don't know if it has something to do with the Solaris/NT relationship.
We can not reproduce it in our labs.
Does anyone have a clue.
Something that will lead us to find a solution?
Thanks in advance.
=============
Problem description:
We are encountering the following behavior in a TCP session between a Solaris based server and a Windows based client:
1. Client sends a request to server.
2. Server sends Ack to Client’s request, then start sending reply packets:
3. Server sends first and second packets of reply (let’s mark them P1 and P2).
4. Client sends acknowledgement for second packet (let’s mark it Ack2)
5. Server sends P3 P4
6. Client sends Ack4
7. Server sends P5 P6
8. Client sends Ack6
9. Server sends P7 P8
10. Client sends Ack8
11. Server sends P10
12. Client sends Ack8
13. Server sends P11 p12
14. Client sends Ack8 Ack8
15. Server sends P14
16. Client sends Ack8
17. …
18. Server sends P20
19. Client sends Ack8
20. Server sends nothing for a period of ~4.5 seconds
21. Server sends P9
22. Client sends Ack20
23. Session continues normally
Step ‘17’ replaces a set of steps where the Server kept sending new packets (P15-P20) and the Client kept sending Ack8 (last accepted packet).
The problem in this session is in step 11 – when the Server skips the ninth packet P9 and doesn’t send it. It appears to be a problem of the Server and not the client of the Network, since both client machine and sniffer machine (two different machines) did not accept the missing packet.
The question is: Is this a known behavior / bug of the Solaris TCP/IP stack? Is there a way to prevent this from happening?
Two more questions related to two parameters that are involved in this session: The number of packets the Server sends before it decides to wait for a missing Ack (in this case P10-P20, meaning 10 packets, on a different machine it was 2 packets only) and The amount of time the Server waits before transmitting the missing packet (in this case in step 20 it was about 4.5 seconds, on a different machine it was 0.5 seconds).
Being able to configure these 2 parameters will help us a lot, in case we can not solve the original problem.
We have witnessed this behavior on both Solaris 2.6 and Solaris 8.
I have a verrrrrry serious problem.
My company product is a server which sits between the Web-server and the internet. It is a kind of proxy.
We installed it at on of our customer which has a Solaris 2.6 Web-server.
Our Server is NT.
Immediately after installing our server, browsing of big files (> 30K) became very slow.
Below is a detailed description of the problem.
We don't know if it has something to do with the Solaris/NT relationship.
We can not reproduce it in our labs.
Does anyone have a clue.
Something that will lead us to find a solution?
Thanks in advance.
=============
Problem description:
We are encountering the following behavior in a TCP session between a Solaris based server and a Windows based client:
1. Client sends a request to server.
2. Server sends Ack to Client’s request, then start sending reply packets:
3. Server sends first and second packets of reply (let’s mark them P1 and P2).
4. Client sends acknowledgement for second packet (let’s mark it Ack2)
5. Server sends P3 P4
6. Client sends Ack4
7. Server sends P5 P6
8. Client sends Ack6
9. Server sends P7 P8
10. Client sends Ack8
11. Server sends P10
12. Client sends Ack8
13. Server sends P11 p12
14. Client sends Ack8 Ack8
15. Server sends P14
16. Client sends Ack8
17. …
18. Server sends P20
19. Client sends Ack8
20. Server sends nothing for a period of ~4.5 seconds
21. Server sends P9
22. Client sends Ack20
23. Session continues normally
Step ‘17’ replaces a set of steps where the Server kept sending new packets (P15-P20) and the Client kept sending Ack8 (last accepted packet).
The problem in this session is in step 11 – when the Server skips the ninth packet P9 and doesn’t send it. It appears to be a problem of the Server and not the client of the Network, since both client machine and sniffer machine (two different machines) did not accept the missing packet.
The question is: Is this a known behavior / bug of the Solaris TCP/IP stack? Is there a way to prevent this from happening?
Two more questions related to two parameters that are involved in this session: The number of packets the Server sends before it decides to wait for a missing Ack (in this case P10-P20, meaning 10 packets, on a different machine it was 2 packets only) and The amount of time the Server waits before transmitting the missing packet (in this case in step 20 it was about 4.5 seconds, on a different machine it was 0.5 seconds).
Being able to configure these 2 parameters will help us a lot, in case we can not solve the original problem.
We have witnessed this behavior on both Solaris 2.6 and Solaris 8.