Session Dissconnection
Posted: 11 Dec 2017, 05:29
I have an Accel-ppp[1] server that deals with a lot concurrent pppoe connections of about 4000 clients. These pppoe connections are generated in a way
that all come in a burst and Accel-ppp server does the pppoe connection hand shake followed by AAA. Accel-ppp server also assigns
HQOS to individual ppp connections. The utility tc (part of iproute2), combined with iptables setting marks on packets and
connections, is used to implement the Hierarchical Token Bucket. This implementation is placed in the /etc/ppp/ip-up script present
in the Accel-ppp server. This script is triggered in the pppd_compat module present in the accel-ppp.conf file at the start of every new connection.
The problem I am facing that when a burst of pppoe request comes in, the Accel-ppp server with the HQOS script is only able to
create a certain number of connections. Than it stops making connections. We have collected data in various configurations e.g.
with HQOS rules and without HQOS rules. Without HQOS rules, all the connections connect to the Accel-ppp server in 1 minute whereas
without HQOS, Accel-ppp is only able to make 15-20% of the total connections and taking a lot of time.
It looks like and TC and iptables are slowing down the connections. Hardware resources where also being monitored while collecting data.
Hardware resources where under utilized.
Hardware specs of server running Accel-ppp:
----------------------------------------------------------------------------------------
Processor |Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz | 32
Memory |64GB total memory | 8 x 8GB
Hard Drive |450GB 10K RPM SAS 2.5” | 8
Network Card |Intel Corporation X520-DA2 (2-ports) | 1
Does anyone have an idea what actually might be the reason for this behavior. What can be the alternative to implement HQOS without TC if it is the problem?
that all come in a burst and Accel-ppp server does the pppoe connection hand shake followed by AAA. Accel-ppp server also assigns
HQOS to individual ppp connections. The utility tc (part of iproute2), combined with iptables setting marks on packets and
connections, is used to implement the Hierarchical Token Bucket. This implementation is placed in the /etc/ppp/ip-up script present
in the Accel-ppp server. This script is triggered in the pppd_compat module present in the accel-ppp.conf file at the start of every new connection.
The problem I am facing that when a burst of pppoe request comes in, the Accel-ppp server with the HQOS script is only able to
create a certain number of connections. Than it stops making connections. We have collected data in various configurations e.g.
with HQOS rules and without HQOS rules. Without HQOS rules, all the connections connect to the Accel-ppp server in 1 minute whereas
without HQOS, Accel-ppp is only able to make 15-20% of the total connections and taking a lot of time.
It looks like and TC and iptables are slowing down the connections. Hardware resources where also being monitored while collecting data.
Hardware resources where under utilized.
Hardware specs of server running Accel-ppp:
----------------------------------------------------------------------------------------
Processor |Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz | 32
Memory |64GB total memory | 8 x 8GB
Hard Drive |450GB 10K RPM SAS 2.5” | 8
Network Card |Intel Corporation X520-DA2 (2-ports) | 1
Does anyone have an idea what actually might be the reason for this behavior. What can be the alternative to implement HQOS without TC if it is the problem?