Forum Discussion

Triggerman_1128's avatar
Triggerman_1128
Icon for Nimbostratus rankNimbostratus
Dec 06, 2006

Performance Metrics

Hi,

 

 

We are currently troubleshooting a performance issues with one of the virtual servers configured on our BIG-IP 9.2.3.

 

 

When going direct, the response time of the web application is (most of the time) less than 1 sec. However, when we go through the BIG IP, the response times is averaging about 10 secs. Both the BIG IP and the Web server are on the same VLAN.

 

 

We are using the following profile:

 

profile tcp test_tcp {

 

defaults from websphere_tcp

 

time wait recycle disable

 

delayed acks disable

 

proxy mss enable

 

selective acks disable

 

limited transmit disable

 

rfc1323 disable

 

slow start disable

 

bandwidth delay disable

 

nagle disable

 

ack on push enable

 

}

 

 

We have captured tcpdumps and logged a call to the F5 support and suggested the following:

 

add the following lines to the bigip.conf:

 

1. route {

 

mtu 1420

 

gateway x.x.x.x

 

}

 

2. Enable "Proxy Maximum Segment"

 

 

It only improve the performance slightly.

 

 

We would like to know if the performance degradation is from the client to the big ip or the big ip to the web server. Can we get these stats using iRules?

 

 

Any help is appreciated.

 

2 Replies

  • What does the websphere_tcp profile look like? Is the client base on the WAN on LAN? We have tuned the default profile somewhat for our WAN clients, the values we have changed are below:

     

     

    proxy mss enable

     

    slow start disable

     

    bandwidth delay disable

     

    nagle disable

     

    proxy buffer low 98304

     

    proxy buffer high 131072

     

    send buffer 65535

     

    recv window 65535

     

     

     

    FYI, enabling bandwidth delay can create instability on your standby unit if you mirror L7 traffic. CR67185 was created for this, unfixed as of this point.

     

     

    Feel free to contact me offline if you need further assistance as the performance issue you are tracking down isn't rules related.
  • citizen_elah,

     

     

    thanks for your reply. Here's what the websphere_tcp profile looks like:

     

    profile tcp websphere_tcp {

     

    defaults from tcp

     

    proxy buffer low 131072

     

    proxy buffer high 131072

     

    send buffer 65535

     

    recv window 65535

     

    }

     

     

     

    We only have one unit (this is in DEV), therefore, we are not mirroring and L7 traffic.

     

     

    We're sure that this is not an iRule related issue but we would like to know if there are any performance metrics that can be used using iRules (logging TCP/HTTP stats) to narrow down possible issues.