CSM and HTTP 1.1 Chunking

[hussainmo]

Hello,

We have an application which uses HTTP chunking which as far as I know supports multiple get requests from the server with one 200 OK. This reduce the overhead and is considered to be more efficient.

The problem we are facing is that with the CSS this application works fine but with the CSM we have to disable this option in the server. Has anyone seen a similar issue or is there anyway the CSM can be made to support chunking?

Thanks

[Gilles Dufour]

http chunking is when the server respond to multiple packets (called chunk) with varying sizes and delimiters to identify beginning and end of the packet.

If you want to send multiple GET and then wait for the server response - http pipelining - this is not supported by the CSM as well as not supported by the CSS.

You can allow this traffic to go through the CSS if you configure L4 or L3 vserver or if you disable persistent rebalance with the command "no persistent rebalance".

Gilles.

[hussainmo]

Hello Gilles,

Thank you for your suggestion. The application works now but can you please elaborate what does the "persistent rebalance" command do and what effect disabling it has on HTTP chunking.

Regards,

Murtaza

[Gilles Dufour]

as I said, the problem is not http chunking.

The CSM supports chunk.

chunk is how the server respond to a client request.

What you were describing is called http pipeling which consists is sending multiple http requests and then wait for the responses.

This is not supported.

CSM expects one request and then one response.

So, by turning off persistent rebalance, to only look at the first request and then not care about the rest of the traffic.

Gilles.

[hussainmo]

Hello,

Ok I got it. So by disabling persistent rebalance HTTP pipelining works because the CSM does not look into each packet in the stream and therefore is unaware of multiple GET requests from the client.

I still have 2 questions though:

1. How does "persistent rebalance" work. I mean what is the difference if its disabled vs. enabled?

2. In the CSS traffic flow I can see the same behaviour i.e. HTTP pipelining but it works with L7 load balancing which should not be the case. I have attached a trace showing this behaviour.

Thanks,

Murtaza

[Gilles Dufour]

if you do not have the command 'no persistent' on the CSS, it would work as it wors on the CSM with 'no persistent rebalance'.

With 'no persistent rebalance' the CSM will make its loadbalancing decision on the first http request then setup a flow in fastpath so that after that the traffic is simply fastswitched [no inspection anymore].

Gilles.

[hussainmo]

Hello,

The CSS also has persistence as I can not see "no persistent" in the content:

content DOCUMENTUM

vip address xxx.xxx.xxx.xxx

protocol tcp

port 80

url "/*"

add service DESAP008_DOCUMENTUM_8181

add service DESAP008_DOCUMENTUM_8282

add service DESAP507_DOCUMENTUM_8181

add service DESAP507_DOCUMENTUM_8282

advanced-balance arrowpoint-cookie

active

BTW I know that on a Cat switch it is the best practice to use hardware switching. Is it the same case for the CSM. In other words should all vservers normally be configured with "no persistent rebalance" for fastpath switching?

Thanks,

Murtaza