Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Mailing Lists -> Oracle-L -> Re: ora-12571 TNS Packet Writer Failure

Re: ora-12571 TNS Packet Writer Failure

From: Jared Still <>
Date: Wed, 1 Nov 2000 15:02:48 -0800 (PST)
Message-Id: <>

On Wed, 1 Nov 2000, Cale, Rick T (Richard) wrote:

> Hi DBAs,
> Oracle 8.0.5 NT 4.0 Net8,
> I have clean install of Oracle client on NT4. I am able to connect to Oracle
> using SQL*Worksheet.
> I cannot connect using plus80 or even do tnsping80 without getting
> ora-12571. I can ping the server
> ok. I do not understand how worksheet can work and not plus80 or tnsping80.
> Any ideas?


I'm repenting of my rather terse previous reply.

Here's some info from MetaLink on this.


Doc ID:


           ORA-12151, ORA-12571 errors on Windows NT   Type:



                                                             Content Type: 
                                                             Creation Date: 
                                                             Last Revision Date: 

  PURPOSE     Give an overview of what to adjust and verify in case of ORA-12151     and ORA-12571 errors

  SCOPE AND APPLICATION     This notes applies to all who are facing intermittent SQL*Net read and     write errors while working on NT.

  ORA-12151 and ORA-12571 errors on Windows NT

  Intermittent SQL*Net TCP/IP read and write errors are sometimes encountered   on NT. The underlying reasons of these errors are a synchronization error   in the TCP/IP layer on NT. To help prevent this kind of errors, a   few things can be adjusted to help the synchronization:

  1. TCP.NODELAY parameter This parameter is to be added in the "PROTOCOL.ORA" file in the "NETWORK\\ADMIN" directory.
     In most cases, TCP/IP info send across the network is buffered 
     till at least a complete network packet can be send. This means
     that in certain cases commands are not issued directly, and kept
     buffered until some other info can be send as well.
     This has the potential to generate timeouts and errors.
     To avoid this, this delay can be switched off.

     tcp.nodelay = yes

  2. Disabling AUTOMATIC_IPC on Clients
     On client PC's, checking for IPC connections is pointless as there
     is never a database installed on them. So, in order to save some 
     time during the connections phase, set AUTOMATIC_IPC=OFF in the 
     "SQLNET.ORA" file.

     If you have a fixed environment, it's best to specify this in the
     "SQLNET.ORA" file. The parameter "NAMES.DIRECTORY_PATH" specifies how
     the TNS resolving takes place.

     By default, if this parameter is not present - the SQL*Net layer
     will first check if there is a Names server anywhere on the network,
     and will afterwards start checking for a "TNSNAMES.ORA" file.

     If you have only a "TNSNAMES.ORA" file, and no Oracle Names installed,
     it is best to specify this parameter to skip the Names searching in
     order to speed up the TNS resolving time.

     The value of this parameter is a comma separated list, with as 
     possible values TNSNAMES (for "TNSNAMES.ORA" searching) and "ONAMES" 
     (for Oracle Names searching).

  4. TCP/IP timeouts on NT
     The default retransmission count on NT is 5, before it detects that
     the network is down. With the value of 5, the actual timeout is
     aproximately 15 seconds.

     This default value can be easily increased to a higher value.
     In the registry, change the following value:

                 TcpMaxDataRetransmissions  REG_DWORD  "number"

     This parameter is not present in the registry by default. This 
     means that the first time, this parameter will need to added to 
     this registry key.

     This parameter can be useful on both client and server. Suggested 
     course of action is to first add this parameter on the machine 
     generating the SQL*Net errors, and if the problem persists, also 
     include the parameter in the registry of the other machine.

  5. TCP/IP keepalive on NT
     KEEPALIVE is an extension to TCP/IP which enables the closing of 
     dead connections which are no longer being used. 

     Problems can occur when the server does not close a connection 
     after a PC client process has disappeared. This typically happens 
     when a PC user switches off or reboots their machine while still 
     connected to Oracle. Note that this is not an Oracle problem, but 
     a limitation of TCP/IP, which has no way of knowing for sure 
     whether a remote connection has disappeared.

     This feature is enabled by default on NT. Problem can occur however
     if the timeout period is too fast for some heavily used or slow
     network. In those conditions, the KEEPALIVE registry value can be
     used to specify a KEEPALIVE value before a connection gets cut.
                 KeepAlive  REG_DWORD  "number"

     A value of '10' is a common value specified for this variable.

     Again, this parameter can be useful on both client and server.
     Start with the machine generating the error, and if needed, also add
     it on the machine on the other side.

  6. TCP/IP timeouts on Windows 95
     The same parameter can also be specified on Windows 95. It has the
     same functionality, only the location of the parameter in the 
     registry is different.

                 TcpMaxDataRetransmissions  REG_DWORD  "number"

     This parameter is not present in the registry by default. This 
     means that the first time, this parameter will need to added to 
     this registry key.

     The purpose and behavior of the parameter is the same on the Windows 95
     and Windows 98, as on the Windows NT platform.

  7. SDU & TDU parameters
     Part of this problem is the sequence of information that gets transmitted.
     If there are disruptions in the sequence, the errors ORA-12151 and 
     ORA-12571 can also appear, alerting the application that not all information
     has been send across the network succesfully.

     The sequence of information is determined by the amount of data the program
     is sending, and the actual size the protocol can send across the network
     at a time. 

     The more data the program wants to send in one 'go', the more sequences and
     transport packet split-ups will have to be made.

     By default, SQL*Net is using an SDU (Session Data Unit) of 2048 bytes (2Kb)
     and a TDU (Transport Data Unit) of 32768 (32Kb) bytes.
     On standard Ethernet connections, with no special modifications made, the 
     SDU is 1500 bytes, while the TDU is 8760 bytes.

     With these values, each data request made by SQL*Net will have to be split
     up in several smaller packets to actually be able to transmit.
     Therefore, to minize the creation the additional packets, it is advised, in
     case of these errors, to synchronize the SDU and TDU parameters at the 
     SQL*Net level with those of the actual network topology used.

     These SDU and TDU parameters have to be specified at both the client and 
     the server configuration files:

        ORCL.WORLD =
          (DESCRIPTION =
            (ADDRESS_LIST = 
              (ADDRESS =(PROTOCOL=TCP)(Host=foobar)(Port=1521))
            (CONNECT_DATA = 
              (SID = ORCL)

          (SID_LIST =
            (SID_DESC =
              (SDU = 1500)
              (TDU = 8760)
              (SID_NAME = ORCL)

     For more information about the SDU and TDU parameter, see Note 44694.1,
     Note 44694.1: SQL*Net Packet Sizes (SDU & TDU Parameters)

  8. Setting a new TDU size on Windows NT
     You can modify the TDU size on NT, via the TcpWindowSize parameter:
Received on Wed Nov 01 2000 - 17:02:48 CST

Original text of this message