[lustre-discuss] Long failover time problem during Lnet bonding

[김성환]

Hi, all.

 

Currently, I have some problem with Lnet bonding configuration.

 

First of all, my server configuration is as below :

 

1 server and 1 client

Lustre version - 2.15.3.

MGT, 1 MDT and 1 OST have on same server. 

Bonding with 2 Infiniband interfaces ( ib0, ib1 ).

Client also configure bonding with Infiniband.

 

And, my problem scenario is as below :

 

Under those configurations, I mounted lustre on client and run benchmarks.

Then, I plugged out ib0 cable from the server.

After that, client I/O was pending 20~40 seconds.

 

I think that It takes too long. 

So, does anyone who has same experience with same configurations?

I wondering that It usually takes that time.

 

If it's not normal, what's the problem?

 

Best regards,

Sunghwan

 

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20230901/f9738386/attachment.htm>