Configure the IBM® Load
Balancer to address dropped audio/video calls when using the IBM Sametime® Connect client.
About this task
This section defines the configuration changes that can be
made to a load balancing component for the Sametime SIP Proxy/Registrar
and Conference Manager nodes to address calls being dropped during
1:1 audio/video and n-way audio/video sessions. Users can be dropped
from an audio/video session within 5 minutes from the time they accept
the invitation to the call, and receive an error message such as:Failure response 408 received in response to invitation to uid=...
Reason is: Unspecified Dial Failure
The Sametime Connect client
registers with the SIP Proxy/Registrar to establish the audio/video
call through an IBM Load Balancer.
The error occurs when the client does not receive a KEEP_ALIVE
response
from the SIP Proxy/Registrar within a specified time and then closes
the connection.
Make the following configuration changes to
the IBM Load Balancer configuration
(.cfg file).
Procedure
- Locate the Load Balancer configuration file in this location
on Linux™: /opt/IBM/WebSphere/Edge/ULB/servers/configurations/dispatcher
- Change the selectionalgorithm value in the port definition
by locating and then removing this definition:
dscontrol port add ipaddress1@port1 selectionalgorithm connection
dscontrol port add ipaddress2@port2 selectionalgorithm connection
Replace
the definition you removed with this value:
dscontrol port add ipaddress1@port1 selectionalgorithm conn+affin
dscontrol port add ipaddress2@port2 selectionalgorithm conn+affin
The
selectionalgorithm modification ensures client affinity for a new
connection, maintaining the client connection to the same server.
The
connection value is based on a connection, whereas conn+affin is based
on a connection as well as the client, so a new connection would use
client affinity. If a client uses Proxy1 for their audio/video session
and then is routed to Proxy2 because of a new connection, issues can
occur, but using conn+affin would maintain the client connection to
the same target server.
- If staletimeout is high, modify the stickytime parameter.
Stickytime is the amount of time the affinity record is kept after
a FIN packet is received and until a new connection is established.
Staletimeout affects the amount of time a connection is maintained.
stickytime=1800
staleTimeout=6400
- From a Linux command
prompt, enter the following commands:
- Determine the interface of the NFA of the Load Balancer
by entering this command:
- Display standard information about the device by entering
this command:
# /sbin/ethtool -k interface_name
- Using ethtool from the command prompt, set these parameters:
# /sbin/ethtool -K interface_name tx off
# /sbin/ethtool -K interface_name rx off
# /sbin/ethtool -K interface_name gro off
# /sbin/ethtool -K interface_name tso off
# /sbin/ethtool -K interface_name gso off
where
interface_name = the interface of the Network Forwarding Address (NFA) of the Load Balancer.
tx: TX checksumming
rx: RX checksumming
gro: generic receive offload
tso: TCP segmentation offload
gso: generic segmentation offload
- When using NAT forwarding, ensure the Maximum Transmission
Unit (MTU) values used on the backend servers are the same, for example,
the SIP Proxy/Registrar and the Load Balancer. The MTU is usually
1500 bytes.
- From the command prompt on the Load Balancer and back-end
servers, SIP Proxy/Registrar and Conference Manager, ensure that the
MTU is set to 1500 bytes by entering this command: