Fixing slow connection problems in Oracle Oracle Database Tips by Donald Burleson

 

Question: We have two database instances on the same server. One was left at 9.2.0.7 and one was upgraded to 10.2.0.3. Connecting externally (sqlplus '/as sysdba') to the 9.2.0.7 database is lightning fast. Connecting externally to the 10.2.0.3 database is very slow, comparatively speaking. This is on an IBM AIX-5L (64-bit) machine. We are using "tnsnames".

The slow sqlplus connection to our recently upgraded database (9.2.0.7 to 10.2.0.3) has been solved by making one change to init.ora. Commenting out the server parameter "_db_block_cache_protect = true" fixed the problem. This parameter does not cause any slow connect problems with our 9.2.0.7 database instances. Any theories as to why "_db_block_cache_protect = "true" is problematic on the 10.2.0.3 database and not the 9.2.0.7?

 

Answer: Please note that connecting via sqlplus " / as sysdba" - is not connecting externally.  If you are doing sqlplus "/ as sysdba" then you are not using your tnsnames.ora file.

 

Slow connectivity can also be related to a memory issue, and bouncing the instance will often fix slow connection issues.  Also, check the PGA for size issues.

 

If you are on 11,2 or earlier, see bug number 7715339, which causes slow logins.  For a permanent fix, (depending on your release or Oracle),apply patch 7715339  or disable event -->event="28401 trace name context forever, level 1"

 

Listener issues and slow connections

 

The listener can sometimes cause slow connections, and you should try connecting diectly, bypassing the listener.  Bouncing the listener (lsnrctl stop, lsnrctl start) can sometimes fix a slow connection issue.  Also see this MOSC note titled "Listener startup / connections hang in 11g" [MOSC ID 803838.1] 

 

External factors and slow connectivity

 

Also note that a slow initial connction may be due to factors outside of Oracle, things like firewalls or entries in the /etc/resolve.conf file. Also, check these files for DNS related connection latency:

eth0 interface (static):

IP : 192.168.1.101
SM : 255.255.255.0
GW: 192.168.1.101
DNS: 192.168.1.101

/etc/hosts :

127.0.0.1 localhost
192.168.1.101 myhost.acme.com myhost (replace myhost with your actual server hostname)

Troubleshooting slow connection problems

 

Slow connectivity can be caused by a variety of issues, and I recommend isolating the connectivity error by trying all possible ways to connect to your instance and then see if there are differences over a db link:

Please read troubleshooting_sql*net_connectivity_errors.

To troubleshoot & diagnose Oracle connectivity problems, I like to start at the simplest, outermost level, and follow this checklist:

1. Check with "ping" using the IP address ("ping 198.11.34.43")

2. Check with "ping" using DNS name (e.g. "tnsping uranus")

3. Try telnet to the IP on port 1521 (telnet 1.2.3.4 1521)

4. Check with "tnsping" using TNS service name

5. Invoke SQL*Plus from the OS command line "sqlplus fred@flintstone".  If this fails, check to ensure that your listener has the flintstone service defined.

6. Sign-on to SQL*Plus and connect with the TNS name ("connect fred/flintstome@service_name")

7. Within SQL*Plus, try a select from table@remote_db_link

This test should isolate the area where the connectivity is slow.

 

Slow connectivity from _db_block_cache_protect

 

As to the use of the undocumented parameter _db_block_cache_protect, you need to contact Oracle Technical Support (MOSC), as Oracle does not support the use of undocumented parameters without their consent.

 

The _db_block_cache_protect parameter is designed to protect database blocks, and should be set to true only when debugging a database RAM memory corruption problem.

 

Steve Adams notes what the _db_block_cache_protect parameter does:

It (_db_block_cache_protect) uses the mprotect() system call to disable access to the block buffers. Every legitimate access must then enable access for the buffer it needs and disable it again afterwards.

This protects you against bugs or hackers corrupting the cache, but it is not really necessary because the log_buffer has similar protection by default.

It also adds about a 30% overhead to system performance. It should only be used by Oracle when diagnosing memory corruption bugs.

Of course, this parameter does nothing on platforms that do not support memory protection, or that support process-based memory protection but have a thread-based Oracle implementation (NT).

 

If you like Oracle tuning, see the book "Oracle Tuning: The Definitive Reference", with pages of tuning tips and scripts.  You can buy it direct from the publisher for 30%-off and get instant access to the code depot of Oracle tuning scripts.