CSS load balancing issue

rv_viji · ‎04-09-2008

Hi,

I'm facing a problem with CSS while load balaning for the web application with two servers.

The application is based on activex..

Basically I have two servers running web application for which I have created VIP in the CSS, user hits the VIP address and they access the application, also we use the sticky thing as the application requires the session persistence.. everything is fine, but the problem starts when one of the server fails...

Assume a user hits the VIP address and access the application, due to the sticky thing his session will be with server A (for eg.), now suddenly the server A fails and in that time the user was doing a transcation and inputting some data and after that he press the submit button on the page, as the server A is down the web page gets refreshed and he has to relogin to the application and redo the whole thing what he was doing in that particular transcation...

Now the application guys are telling this problem should not happen as the CSS should be able to take care of the session getting reestablished to the other server B during the server A failure...

Can someone through some lights on this... I'm bit confused now... as what I understand is that the webpage gets refreshed during a server failure because the tcp session id will get changed and the server B will not accept the same tcp session so it reinitiates the new session...

Is my understanding right?? or is there something which we can do on the CSS to avoid this problem...

Regards

Vijay.

Gilles Dufour · ‎04-09-2008

Your application guys are wrong.

The CSS will reestablish a session with the new server, but since the server does not know this client it does request a new login.

They would need to have some inter-servers synchronization for this to work and this is not very common.

Gilles.

rv_viji · ‎04-09-2008

Hi Gilles,

Thanks for the clarification.

I have two more issues too...

1. The load balancing of the application between the two servers are not even. Actually the traffic from the users keep hitting only one server, I understand the point of sticky method used in our case, but even atleast the connection from another client machine should go to the other server,but it is not the case... traffic from all the clients goes to only one server..

what could be the possible reason for the same...

My config is as below...

service SERVER-1

port 80

protocol tcp

keepalive port 80

keepalive type tcp

redundant-index 4

ip address 10.6.223.87

active

service SERVER-2

port 80

protocol tcp

keepalive port 80

keepalive type tcp

ip address 10.6.223.77

redundant-index 5

active

owner WEB

content WEB

add service SERVER-1

add service SERVER-2

redundant-index 104

vip address 10.6.223.78

protocol tcp

port 80

url "/webretrieve*"

advanced-balance sticky-srcip

active

2. Slow response of the application when users access application through VIP address(CSS), what can be done further in the configuration to improve the performance?? or any thing else I can do...

Regards

Gilles Dufour · ‎04-10-2008

THe problem with stickyness is if one server goes down, all connections are sent to the other server and a sticky entry is created.

So, even when the down server comes back up, because of the sticky entry, all traffic continue to go to the other server.

After a server going down, you should clear the sticky table at a time it is less impacting.

If you have frequent server failures, you should fix the servers.

Gilles.

rv_viji · ‎04-11-2008

Hi Gilles,

The problem stated in #1 is not happening during the server failures, I'm talking about a situation where both the servers are up and working fine, and when the users hit the VIP address for accessing the application we see the traffic is being sent to only one server and not getting load balanced evenly.

Regards

Vijay

Gilles Dufour · ‎04-14-2008

Vijay,

with stickyness this is usually the case.

If you take a basic example of 2 clients.

One sent to server A and the other sent to server B.

If client 1 opens 10 connections and client 2 only 1, you will see 10 connections on server A and only 1 on server B.

All this is normal with stickyness.

Gilles.

damico_nick · ‎04-17-2008

If everyone is going through a proxy this could explain what you are seeing. Remember, your sticky decision is being made by src IP. If everyone is using a proxy, then everyone is seen as the same src IP.

Solution would be to bypass the proxy either by a bypass list if using wccp or IE exlusion if using explicit proxy.

Hope this helps.

Nick

agiaccone · ‎04-17-2008

As Nick says a proxy could explain the behavior you described.

In this case changing the sticky method to arrowpoint-cookie should resolve, as decisions would be made by a cookie inserted on the client and not bye the source ip (i.e. the proxy).

You can look here for reference :

http://www.cisco.com/en/US/docs/app_ntwk_services/data_center_app_services/css11500series/v8.20/configuration/content_lb/guide/Sticky.html

HTH

Alberto

lamav · ‎04-17-2008

Vijay,

with stickyness this is usually the case.

If you take a basic example of 2 clients.

One sent to server A and the other sent to server B.

Gilles, thats just the point it seems: traffic is NOT making it to server B in the first place.

Vijay:

Are you indeed using a proxy as Nick and Alberto suggest?

Thanks

Victor