Results 1 to 3 of 3
Hi everybody,
I'm not sure if the following setup causes problems:
All machines run on RHEL4 (2.6.9 kernel).
I have a disk pool with two NICs connected to a switch.
...
- 03-21-2008 #1Just Joined!
- Join Date
- Mar 2008
- Posts
- 1
nfs - two nics same subnet
Hi everybody,
I'm not sure if the following setup causes problems:
All machines run on RHEL4 (2.6.9 kernel).
I have a disk pool with two NICs connected to a switch.
The two NICs got IPs in the same subnet (see below).
Then I got two machines (WN) with one NIC each also
connected to the switch. These machines, that live in
the same subnet as the disk pool, import
disk space from the disk pool via NFS:
First WN (192.168.1.21) via the disk pool interface (192.168.1.10)
Second WN (192.168.1.22) via the disk pool interface (192.168.1.11)
SUBNET:
192.168.1.XX/255.255.255.0
____________
| |
| Disk Pool |
| 10 11 | = XX
|____________|
___|____|___
| |
| Switch |
|____________|
_|__ __|_
| 21 | | 22 | = XX
| WN | | WN |
|____| |____|
The reason for this is to have two simultanious GBit connections
from the disk pool to the WNs.
Is there a routing problem, because of the the disk pool NICs
being in the same subnet?
Does it make a difference using TCP or UDP for the NFS exports?
Thanks and cheers,
RS
- 03-25-2008 #2
First are you are not going to use a gig connection 100% of the time. From what you have written this is just for disk space. I am going to assume that the disks are not going to be written to continuously. If this is the case then one connection from the linux box to the switch is more then enough.
Connecting 2 Nic to the same switch will cause you issues and is not needed from what I am assuming here.
- 03-25-2008 #3Linux Guru
- Join Date
- Nov 2007
- Posts
- 1,679
2 NIC's in the same subnet *may* cause problems depending on the application in use - packet reordering, conversations that switch between source MAC's, etc.
If you actually want the bandwidth/redundancy of 2 NIC's, you may want to look at NIC bonding.
No - there is no routing involved. The NIC's are all on the same network. The OS will choose a NIC to use. Since they're both on the same network, it can select either one.Is there a routing problem, because of the the disk pool NICs
being in the same subnet?
This is a performance question. *IF* your switch and network cables are reliable, using UDP should give you better performance - being "connectionless" means less overhead, at the cost of reliability/being sure the data got there.Does it make a difference using TCP or UDP for the NFS exports?


Reply With Quote
