Explanation of Oracle Direct NFS network traffic test (1)

In terms of databases, NFS can also be applied. Today we will focus on Oracle Direct NFS. In this article, we have explained in detail the test process and specific configuration methods, and hope to help you. Then, we tested the Oracle Direct NFS function and found that ORACLE Direct NFS improves IO concurrency by establishing multiple TCP connections to the NFS Server. We mentioned earlier that the reason for the low I/O capability of NFS is that the operations from the NFS client to the NFS Server are serial, and only one connection is established from the normal NFS client to the NFS Server, in addition, the latter request can be processed only after the previous request is processed. In this way, the random read IO will fail. Oracle Directd NFS and NFS Server establish multiple TCP connections, and the processing can be performed concurrently. In theory, the NFS performance can be greatly improved.

The actual Direct NFS reading was very fast, and the actual test reached 400 Mbytes/s, basically no bottleneck was found. However, when writing, it was found to be slow. when inserting data, the write traffic is only about 3.4 Mbytes/s. The reason why the write speed is so slow is unknown. It is estimated that the Linux NFS Server does not work well with Oracle Direct NFS. When rman is used for backup, if the backup path is in the path specified by Direct NFS, it will automatically go to Direct NFS mode.

Test process:

First modify the odm database and start the odm database that supports Direct nfs:

[oracle@nfs_client lib]$ ls -l *odm*
-rw-r–r– 1 oracle oinstall 54764 Sep 11  2008 libnfsodm11.so
lrwxrwxrwx 1 oracle oinstall    12 Jul  8 18:55 libodm11.so -> libodmd11.so
-rw-r–r– 1 oracle oinstall 12755 Sep 11  2008 libodmd11.so
[oracle@nfs_client lib]$ rm libodm11.so
[oracle@nfs_client lib]$ ln -s libnfsodm11.so libodm11.so

Share a directory on the nfs server machine. In order to use the hard disk without becoming an I/O bottleneck, use eight disks as a RAID 0 and then an ext3 file system as the nfs Server output:

mdadm -C /dev/md0 –level raid0 -c 8 -n 8 /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf /dev/sdg /dev/sdh /dev/sdi
mkfs -t ext3 /dev/md0
mount /dev/md0 /nfs

Then configure in/etc/exportfs:

/nfs 192.168.172.132(rw,no_root_squash,insecure)
service nfs restart

On the Database Host (nfs client ):

[oracle@nfs_client dbs]$ cat oranfstab
server: node_data1
path: 192.168.172.128
export: /nfs mount: /opt/oracle/oradata/nfs
mount -t nfs 192.168.172.128:/nfs /opt/oracle/oradata/nfs

The two machines are connected by a 10-Gigabit Nic, and the tested network speed can reach 800 Mbytes/s or higher.