OVH Community, your new community space.

work tonight: split vss-1
09-02-2010, 04:14 AM
Good evening,

completed (with great difficulty and a problem on IP failover on the HG).

09-01-2010, 09:58 PM

1 month ago we "split" the vss-2 (double frame) to a configuration in a single chassis. This has fixed all the problems we had with the vss. The BGP works quickly and does not further saturate the CPU and more ARP problems.

We will "split" vss-1 router tonight at midnight. It will take about 30 minutes to complete. This is a configuration change and a total reboot of the router to change all the configuration completely. During this 30 minute period, the servers managed by the vss-1 will be cut.

More information:

After the vss-1, there will be more than the last vss-3. As a result, we do not have more problems.

2 years ago, we made the wrong choice of configuration and we waited, in vain, to fix our problems. By pretending that the past "with the means at hand", all the problems still to be determined, will be resolved one by one without waiting. We still have two weeks of work.

Otherwise, we will not use the Nexus 7016 as we originally planned. We ordered one to do the tests and what we received was not what we expected. There are RAM limitations that prevent us from running the BGP properly, the Cloud Computing functions are undeveloped and different bugs in the development for example of the proxy-ARP with HSRP with forced mac , no proxy-arp ipv6 etc.. In short, the hardware is not ready for the moment to work in L3. In short, we lost time. Lots of time. We had orders to Juniper and Brocade for similar equipment and ASR 9000. All this to make the tests (for now) and see firsthand the operation and bugs ...

It challenges our plans? Oddly, no. The routers in use today function properly to a certain level. The idea is not to exceed this limit rather than seeking a new material with which one could pass it. And so instead of upgrading the routers, we will implement something new in parallel. Simple. The backbone will be "less centralised" and therefore "more distributed". Divide, divide and divide to make 2 or 3 parallel backbones, each of 500Gbps of capacity and each operating almost independently of each other.

The other network in progress:

BGP route reflector:
QoS Problem on the rack switches:
Protection against attacks in UDP:

All the best,