Switching

Hello,

We got a Virtual-chassis that is made from 6 switches, 4 EX4550 and 2 EX4200

2 EX4550 and 1 EX4200 at our main site.

2 EX4550 and 1 EX4200 at our backup site.

We got 2 10G connection between the site.

We followed Juniper recommandation and didn't disable the split detection.

We configured the master at our main site, and the backup at our backup site.

When we shutdown the switches at the backup site the main site went into linecard role, and our entire communication went down.

We followed Juniper documentation and we discovered that:

The "Enabled" is for a split-active detection.

Is there a way to conifgure "MAD" at Juniper that the discovery of "split-brain" will be more accurate?

Is there any solution to have a split-detection but when our backup site shutdown not our entire network fails?

Thanks,

Omer Shtivi

Split detection really only needs to enabled for 2 switch configurations.  It appears to me you have a 6 switch mixed-mode VC that is connected via eVCP - multiple 10G connections between local and some geographically remote site.  Was the VC set up with pre-provsioning?  If not I would recommend this, as I suspect what happened was that the 2 x EX4550s on one side, the side you powered down, had both of its EX4550s with role of RE-Master and RE-Backup.  Therefore when you powered down that side, remote side all remained in role of LC, with no RE, and therefore no forwarding.  What you want to do is force RE-master on one side EX4550, and RE-backup on other side, either via pre-provisioning, or via VC Master priority configuration:

http://www.juniper.net/techpubs/en_US/junos12.1/topics/task/configuration/virtual-chassis-ex4200-mastership-cli.html

Otherwise for this configuration split no-split-detect needs to be disabled.

This should solve your situation from happening again.  This needs to be an added consideration for any remote VC design using eVCP connections.

Hi,

Thanks for the response

Actually juniper recommanded enabling split detetction with more then two switches:

Best Practice: We recommend that you disable split detection for a two-member MX Series Virtual Chassis configuration if you think the backup router is more likely to fail than the Virtual Chassis port interfaces to the backup router. Configuring redundant Virtual Chassis ports on different line cards in each member router reduces the likelihood that all Virtual Chassis port interfaces to the backup router can fail.

Is there a way to run both split-detection and avoiding the scenrio when the backup shutdown the entire VC stop working?

FYI, the recommendation you quoted was for MX VC which by default is always only 2 switches, but potentially 4 x REs.  Totally different situation than your mixed mode VC with EX4550/EX4200.

Split detect (no-split-detect disabled) and/or proper configuration should ensure you do not have a repeat, but what you will probably get instead is potential for split brain - that is both sides active as 2 x standalone VCs.  Then when re-connected, the 2 x VCs will need to negotiate RE-M and RE-B.

So there is no way to use split detection at QFX/EX in a six switches secnario?

Yes just configure no-split-detect disbled => this is double negative which enabled split-detect.  yes I know, confusing!