Socket errors

[John Peace]

HPUX 11.0
HP K410

Running a Progress Database with aprox 150 concurrent users at any one time. Users are using a GUI connection. One certain customer base is getting constant timeout errors. Error message is as follow:

"Error reading socket n=# ret=1 errno=10054"

This is on the users machine. The errno.h listing states the connection was reset by peer. My understanding is "somehow" the server (the HP) and the client have lost connection and the client has cleared the connection for their TCP stack. When the server sends some packets back the client is not aware of the connection and resets it. WE know for a fact the connection between this set of clients and us in terrible, but no one wants to "fix" it. My questions:
1. Do I have this backwards since the errno is from the NT and not the HP. Is the HP resetting the connection?
2. Is there a tcp "ndd" setting I can change to increase the timeout time? Maybe the
tcp_rexmit_interval_max....currently 1 minute
tcp_rexmit_interval_min....currently 4 sec.
3. How will it affect my other users? Only 5 of 150 are at the "bad" location.
4. Is there a good book that tells more about tcp keepalive and other ndd setting? I think I know, but I wonder.....

[Berlene Herren]

This is normally caused as a result of slightly different default behaviour of sockets under UNIX to those of NT. The problem is very well described in the Microsoft Knowledge database under PSS ID Number: Q105794

Workaround
Add the registry entry: TcpKeepCnt, or modify it if you already have one in the following path: HKEY_LOCAL_MACHINESYSTEMCurrentControlSetServicesTcpipParameters You should set the value to zero. For example,TcpKeepCnt:REG_DWORD:0. NOTE: You need to reboot Windows NT after making this change.

let's try this first, just with the 5 users affected.

Regards,
Berlene

[Albert E. Whale, CISSP · Accepted solution]

John,

I take it that you've already performed the ndd -h and gotten the help information?

First off, I think that you know that it will be significantly easier to fix the broken connections, rather than to tune your environment for the exception. But that's your choice, and I realize that as an Admin, your hands may be tied on this, so here goes.

1. Do I have this backwards since the errno is from the NT and not the HP. Is the HP resetting the connection? <- Well that depends, who was trying to communicate? This is almost impossible to track down, suffice it to say that the Network Connection is poor, does it matter who was talking last?

2. Is there a tcp "ndd" setting I can change to increase the timeout time? <- I believe that what you want to do is to decrease the Maxiumum retry time. Or increase the number of Keepalives (i.e. tcp_keepalive_detached_interval). You must be careful though, becasue whatever you do here, is certainly going to increase the Network Overhead associated with the entire environment.

3. How will it affect my other users? Only 5of 150 are at the "bad" location. <- This is a loaded questions because it depends on many factors. Suffice it to say that the overall network traffic load will increase with the use of additional keepalives (if that's what you actually use).

One thing that I haven't heard yet but needs asked, does the network connection Terminate, or just quit responding? Perhaps if the Network connection does not terminate (and subsiqnetly re-initiate), increasing the timeout value will help, and will not impede network performance.

4. Is there a good book that tells more about tcp keepalive and other ndd setting?

Try this link:

http://docs.hp.com/dynaweb/hpux11/@Generic__CollectionView?DwebQuery=ndd

Hope that helps you!

[Paul Hite]

In reference to a good book, I would like to suggest "TCP/IP Illustrated Volume 1" by the late Rich Stevens. This is the best book on TCP/IP that I have ever read.

[Berlene Herren]

I have to agree with Paul on book selection - all three are great!

Berlene

[Shailesh V. Marathe]

Let me first thank you all for your replies to this question posted by John.
Actually I am also facing a similar problem. I have 64-bit ver.11.00 Hp-servers K360.
These are my application servers which are connected to a Main Database Server.
The problem is when I ping one of the server from the Win-95 PCs I get frequent request timed out messages whereas at the same time I am to ping from other Servers to this Server (in short Unix to Unix communication is Ok but Win-95 to Unix gets broken). My BaaN clients are loaded on Win-95 Pcs and Database is Informix.
I have tried out the settings recommended by Berlene in one of my NT workstation and kept it in ping state. Still I get request timed out.
I have checked all my cables & 3-Com Switch settings and are Ok.
Please can anyone give any idea what exactly is happening.
Thanks in advance.

Shailesh

[Domingo Muñoz]

Hi,

I'm having a similar problem.

We have a cluster with 2 node and when the services is move to the secudary node to the client appears the error 10054 .

After I read our messages, I have added and modified regedit of the client and I have increased the tcp_rexmit_interval_max parameter to 3 minutes, but continue the same problem.

Is there somebody that I can help with this problem?

[IT Response]

Hi John,

Just a quick note. I recently ran in to a similar error message with the CallView application using a Progress Database. Clients were unable to retrieve data and timing out.
No errors in syslog or nettl. Not sure on the number of users though. You might want to check your Streams and ARPA patch level.
Current patches resolved issue in this case. This was on a gigabit card and the driver was updated too. Did not remove driver to check patches individually.

Good Luck,
Chris

[paulette brown]

Im experincing the same problem, and was wondering if there is a solution. I have more users affected by this timeout error. very annoying. any help would be greatly appreciated.