1 Reply Latest reply: Oct 24, 2017 3:08 AM by rgirdlestone . RSS

different inpath (PBR) for same session

Steven Wong

Hi there,

 

I have a virtual in-path (PBR) deployed in all locations lately. It's working great but I started seeing problems since then.

 

On the network edge, we used to have two layer 3 switches acting both top of rack switch for the servers in the rack and also as an edge router for traffic over WAN links. Usually we have two WAN links with one pointing to the DC in east/south and the other pointing to the west/north. On Riverbed, we have wan0_0 on one edge while wan0_1 on the next edge.

 

Before the last weekend, we were still running RiOS 8.5.3. In cases the client tries to connect to server via local edge and return traffic is via the next edge, the riverbed cannot establish a session because of AR. (forward path is wan0_0 but return path is wan0_1).

 

After upgrading to RiOS 9.6.1 last weekend, I found those unoptimized sessions (due to AR) is being optimized now.

 

I couldn't figure out how this is done and wasn't aware of special feature on 9.6.1 that can optimize the traffic even the inpath is different.

 

Can someone give me a shed of light if it's a known feature to avoid AR if the session is on the different inpaths of the same Riverbed?

 

Just want to share that our major traffic is kafka cluster streaming, mysql and general smb and the optimization ratio is great for kafka where application compression is enabled and we still gain around 80% data reduction.

 

Thanks,
Steven

  • Re: different inpath (PBR) for same session
    rgirdlestone .

    Hi

     

    I am not 100% certain, as I have no steelhead old enough to test.  However, I do remember that using PBR, only the lowest in-path interface number (in-path0_0) would optimize, unless you issue the hidden command "in-path oop all-ports enable".  In which case it would allow other in-paths as well.  I have gone back as far as I can in my lab (8.6) and the command is no longer there, hidden or otherwise.  I can only summit that no all ports are enabled for PBR regardless.

     

    As I said, I am not 100% sure, without testing it, but it may be a steer for you.

    Can I ask if there is a problem with it, or just an observation of the new behavior?

     

    regards

     

    Rupert