Getting multi-path direct NFS to work on linux_x86_64
657156Aug 28 2008 — edited Feb 25 2009I have linux box running 11gR1. The box has 2 gigbit NIC's installed and running (each with its own IP). We run oracle over NFS on a netapp. The netapp has 8 gigbit cards in it. I would like to use directNFS to get ~2gigbit net throughput, but I cant seem to get it to ever use the other card. There is a metalink ticket with "dnfs load balancing does not work on linux" but no details that I can see, an no resolution. Its not clear if that is the same problem I am having.
When running the io calibration, I consistently get:
Maximum IOPS: 14096
Maximum MBPS: 107
I believe I should be getting closer to 200MBPS. Also, I run the linux system monitor and I see all traffic going over eth0, while eth1 does barely anything.
I believe eth1 is installed and configured correctly. I can ssh to the host using eth1 IP, I can ping it, there is some (very little) traffic on it. The /etc/sysconfig stuff looks right, ie:
cat /etc/sysconfig/network-scripts/ifcfg-eth1
DEVICE=eth1
BOOTPROTO=static
HWADDR=00:14:4F:0D:5B:75
IPADDR=14.1.72.130
NETMASK=255.255.248.0
ONBOOT=yes
#TYPE=Ethernet
My oranfstab looks like this. I am defining 2 paths to the netapp.
sh> cat $ORACLE_HOME/dbs/oranfstab
server: paz
path: 192.168.6.1
path: 192.132.2.117
export: /vol/db1 mount: /prediction/databases
When I startup the database, the relevant elements in alert log. Note that only 1 of the paths is in the log, not sure if thats normal:
<txt>Direct NFS: mount complete dir /vol/vol1/local/Linux2.6.9-55.ELsmp/pkg/oracle on pez path pez mntport 4046 nfsport 2049
<txt>Direct NFS: mount complete dir /vol/db1 on paz path 192.168.6.1 mntport 4046 nfsport 2049
Then I query some of the dnfs tables and it looks like everything is setup, and both paths are being used.
SQL> select svrname,path,sum(sends),sum(recvs) from v$dnfs_channels group by svrname,path;
SVRNAME PATH SUM(SENDS) SUM(RECVS)
-------------------- -------------------------------------------------- ---------- ----------
paz 192.168.6.1 36835 229280
paz 192.132.2.117 40653 265684
pez pez 128 256
All this, but no traffic over eth1 and thus no net bandwidth improvement.
ideas greatly appreciated.