Nathaniel Rutman
2007-Jan-23 11:58 UTC
[Lustre-discuss] Re: [Lustre-devel] mount cmd fails on OSS
Snider, Tim wrote:> Nathan, > Thanks for the response. Making more progress. The ''depmod'' command > wasn''t run - so all the modules got errors during the load. Once I did > that I was able to do a mount on the OST and get the "Is MGS running" > error message - this is good - indicated something was happening. > For some reason the option parsing routines in my kernel version don''t > work as Lustre expects. Don''t know (really care) why - I''ve hacked .c > files in utils to get mkfs.lustre to work. I expect the same will be > done with the mount code. > > BTW - Have y''all published definitions for the new types? > MGS - config. mgmt. server. What''s the ''G'' stand for in MGS and > MGC? >ManaGement Server. We like TLAs I guess.> MGC - Config. mgmt. client? > MDT - Meta Data Target? does this replace MDS? >Yes, the (old) MDS was unfortunately misnamed back in the day, and we''ve been stuck with it. We are planning to fix this (swap the old MDS and MDT names) but for now, it just shows up as the 1.6 device name, and obd type names (that show up in /proc/fs/lustre/devices column 2) retain their old wrongness. The way it should be, and the way we talk about it in the 1.6 documentation: T=Target (disk), S=Server (node). There can be multiple MDTs on an MDS. The MDS is like the OSS; it handles comms and passes messages along to local **Ts.> OST - Object Storage Target. - remains the same. >Yes, since it represents a single disk. We want to fix the obd type name to become "ost" instead of "obdfilter" here too.