MULTIPLE INTERFACES TO THE SAME NETWORK IN AIX.
ITEM: RTA000036222
QUESTION:
When there are multiple ethernet interfaces configured for the
same subnet over separate adapters, under what conditions will the
adapters be utilized?
Example
A PC ftp's to AIX using the address of the second adapter interface
and then does a GET. Will the file transfer go over the second
adapter or will it use the first route in the routing table that
matches the network of the destination?
The objective of the customer is to load balance across the adapters
by having users access the AIX machine over different interfaces.
---------- ---------- ---------- --------- ---------- ----------
A: There are two ways to split up traffic over multiple adapters:
(1) have different users specifically telnet/ftp into one or the other
adapter card, or (2) use one adapter for incoming traffic and another
adapter for outgoing traffic.
The only way to split up network traffic over two (or more) adapters
is to specifically send traffic to the desired adapter. Each adapter
has to have a separate IP address. If you wish to split up traffic
between the two adapters, some of the traffic would have to be
specifically addressed to one of the IP addresses and the rest of the
traffic would have to be addressed to the other IP address.
For example, lets take the case of a RISC System/6000 with two
ethernet cards on the same network. Adapter 'A' has an IP address
127.1.1.1 and adapter 'B' has address 127.1.1.2. There are ten users
who wish to use telnet. If five of those users telnet into address
127.1.1.1 and the other five telnet into address 127.1.1.2, the
network traffic will be almost evenly distributed over the two
adapters.
Now let's take your example. You are correct that when an FTP session
is started, the first route in the routing table that matches the
network of the destination will be used. Once the connection has been
established between one specific adapter on the client machine and a
specific adapter on the server machine, there is no way for traffic
to be routed over another adapter in the server to distribute the load
on that adapter. Therefore, there are two conditions under which one
of the adapters will be utilized: (1) a user ftp's into that adapter,
(2) the route table points to that adapter.
In summary, load balancing between two network adapter cards cannot
be done on a RISC System/6000. As a side note, the network generally
proves to be a greater bottleneck than the network adapter cards.
Hence any attempt to split traffic amongst two adapter cards may prove
more confusing to the end users than it is ultimately worth.
---------- ---------- ---------- --------- ---------- ----------
QUESTION:
Now how about NFS? If you execute a mount using a server address
of (9.10.1.1) will the data returned from the NFS server to the
client come over the 9.10.1.1 address or the first address in the
routing table for the 9.10.1 network (ex 9.10.1.2)?
What about an application written using UDP over sockets? Will all
data be passed over the adapter that the connect was received over? Or
will it pass over the first adapter in the routing table going to
that network?
---------- ---------- ---------- --------- ---------- ----------
A: Below is a diagram which may facilitate this discussion.
+------+------+------+------+------+------+
| | | | | | |
| XWin | SMTP | Tlnt | FTP | SNMP | NFS | Application Layer
| | | | | | |
+------+------+------+------+------+------+
| | |
| TCP | UDP | Transport Layer
| | |
+---------------------------+------+-----++-----+
| | | |
| IP and ICMP | ARP | RARP | Internet Layer
| | | |
+----------------------------------+-----+------+
I will now address your questions in the order presented.
1. NFS.
When you mount to address 9.10.1.1, the server will return all
data using the first route in the routing table which matches
the IP address of the client. Thus if the first address in the
routing table for the 9.10.1 network was interface 9.10.1.2, all
return traffic will be returned through the 9.10.1.2 adapter.
Therefore, in this case, input would be received from the 9.10.1.1
interface and output would be sent through the 9.10.1.2 interface.
2. Socket application using UDP.
This scenario is the same as the NFS scenario above. The diagram
at the top of this response helps explains why. Determining which
interface is to be used to send data to a machine is done by the
Internet Layer. NFS and your socket application are at the
Application Layer. Thus there is no real distinction between the
two cases as far as addressing are concerned.
In summary, the applications (NFS, socket based programs, telnet,
ftp, ftpd) run at the Application Layer. Routing is done by the
Internet Layer.
To summarize, all TCP or UDP traffic is routed according to the
routing tables. Therefore, it is possible (although not necessary) to
receive all TCP/UDP traffic on one interface and send outbound traffic
out another interface.
---------- ---------- ---------- --------- ---------- ----------
This item was created from library item Q647023 CMRDP
Additional search words:
AIX CMRDP COMMUNICATIO INTERFACES IX JAN94 MULTIPLE NETWORK
NETWORKING OZNEW RISCSYSTEM RISCTCP SAME SOFTWARE TCPIP TP
WWQA: ITEM: RTA000036222 ITEM: RTA000036222
Dated: 04/1996 Category: RISCTCP
This HTML file was generated 99/06/24~12:43:13
Comments or suggestions?
Contact us