Network Tuning (HP-UX) for Oracle DataGuard
We would like to do some Network Throughout and Database Throughput tests to findout which is a better option between, HP Continuous Acess or Oracle DataGuard technologies. We are a major bank and we have about 10g R2 RAC (10.2.0.3) Databases using CFS and running on HP-UX 11i V2 Itanium in our Primary Site and we will be having 10 RAC DataGuards for these Primary RAC Databases.
So, In our basic tests using LGWR SYNC/ASYNC and ARCH mode of LOG Transport methods, we found a major time difference in completeing a huge transaction. So, we would like to do some Network Tuning like increase TCP/IP Send Buffer Size etc. Please let me know if anyone did the similar exercises. Also, I want to know how to do that on HP-UX 11.23 Itanium.
1. To monitor network statistcis and retransmitted data,
executet the following command:
netstat -s -p tcp
If the num of retransmitted data packets is high,
then your network is not having enough bandwidth.
2. The next command show all the tcp related parameters in 11.23
ndd -get /dev/tcp ?
3. The parameters important for peformance are:
Thanks for your info. I know the commands to check the parameters for the network protocols using "ndd"
I was actually looking for the changes to the Send and Receive Buffer Sizes:
(Obtained using "ndd -h supported")
tcp_recv_hiwater_max - Upper bound on TCP receive buffer size
tcp_xmit_hiwater_max - Upper bound on TCP send buffer size
Are you sure that we have to modify:
Also, what are the maximum supported values and what are the other things they affect?
Both parameters can be set to minimum of 512KB (524288 bytes).
I do not remember the maximum values. Try with 1MB and do some benchmark with various values.
Last edited by tamilselvan; 05-12-2008 at 07:01 AM.
Thanks for your info. The following are the current values that I took from my system:
$ ndd -get /dev/tcp tcp_recv_hiwater_def
$ ndd -get /dev/tcp tcp_recv_hiwater_max
$ ndd -get /dev/tcp tcp_xmit_hiwater_max
$ ndd -get /dev/tcp tcp_conn_request_max
$ ndd -get /dev/tcp tcp_ip_abort_interval
$ ndd -get /dev/tcp tcp_keepalive_interval
Any Comments? Also, I wanted to know just setting the Buffer Sizes will do? I mean are there any dependent parameters which need to be set to avoid congestion or buffer overflow kind of situations?
Click Here to Expand Forum to Full Width