This group is locked. No changes can be made to the group while it is locked.
Re: How to compute the routing path between two hosts in OpenDayLight?
Amit Mandke (ammandke) <ammandke@...>
Richard and Junzhuo,
I would highly encourage you to submit your enhancements to l2switch. Please start contributing the project. Submit your proposal for Beryllium release.
From: <yang.r.yang@...> on behalf of "Y. Richard Yang" <yry@...>
Date: Sunday, August 9, 2015 at 7:01 AM
To: Junzhuo Wang <wangjunzhuo200@...>
Cc: 陈明明 <mingmingminne@...>, Amit Mandke <ammandke@...>, Evan Zeller <evanrzeller@...>, André Martins <aanm90@...>, "Tony Tkacik -X (ttkacik - PANTHEON TECHNOLOGIES at Cisco)" <ttkacik@...>, "l2switch-dev@..." <l2switch-dev@...>, Shu Dong <dongs2011@...>, "linxiao9292@..." <linxiao9292@...>, Jensen Zhang <jingxuan.n.zhang@...>, "gaok12@..." <gaok12@...>, "alto-dev@..." <alto-dev@...>, Andreas Voellmy <andreas.voellmy@...>, Shigang Zhu <szhu@...>
Subject: Re: How to compute the routing path between two hosts in OpenDayLight?
To add a bit, according to my understanding:
- current l2switch uses flooding based on a single, shared spanning. Hence, assume a switch with k neighbors in the spanning tree, it needs only k rules in its flow table: for each incoming neighbor, flood to the other k-1 neighbors. If we use the number of links traversed as cost when computing ECS, this routing scheme is not useful, because the cost for each arc-dst pair is the same.
- As a first step, we develop l2switch-plus, which adds higher priority flow rules: for a given src, dst MAC pairs, we find the unicast path from the tree, and add <src-Mac, dst-Mac> entries to the flow tables. Note that this may cause flow table explosion, which we can look into:
* We identify compression.
* Another idea is that when computing ECS, we do not insert the Mac rules. They are triggered by packet-in only.
* Yet another idea is that ECS will insert the rules but with a relatively short timeout. Hence those rules not used by app will timeout quickly.
- In the next step, we use the ip-mac binding table to convert an l2 path to an l3 path.
In the long run, the preceding is not complete. For example, what if GBP reroutes a flow through middle boxes. The preceding derivation based on l2switch routing is not correct. We need to look into the composition approach of ODL.
On Sunday, August 9, 2015, Junzhuo Wang <wangjunzhuo200@...> wrote: