Print

Print


On 6/15/19 1:27 AM, Andrew Hanushevsky wrote:
> Hi Adrian,
Hi!

> On Fri, 14 Jun 2019, Adrian Sevcenco wrote:
> 
>> on the dfs topic with central lookup, how is the redirector aware about
>> the location of the data? i can make sure that the mountpoint is 
>> identical over all servers (redirector and xrootd data servers alike)
>> but should i also set up on the redirector things like oss.space
>> and for ALICE case the specific authorization and file name translation
>> (ofs.authorize ofs.authlib oss.namelib) ?
> Presumably, the redirector has access to the DFS. Since in a DFS setup 
> all servers have the same file then if the redirector finds the file in 
> the DFS it knows that all the servers must have the same file. In this 
> case, the DFS paths must be exported the same way in the redirector as 
> in the servers and the same Name2Name must be used. I don't see why you 
> need oss.space (other than token assignment) if you are using a DFS. The 
> simplest solution is to use the same config file for everyone with 
> minimal exceptions for the redirector using if-else-fi. There is no need 
> to do authorization (and it would fail if you did) for the redirector.
ok, thanks a lot for all clarifications! i forwarded these details to
my ALICE italian colleagues

>> hmm, while i use cms.sched i did not use cms.perf
>> and i see that cms.perf have no pgm default even if there is a 
>> provided script /usr/share/xrootd/utils/XrdOlbMonPerf
> Correct. Thet's the default script we have. You may need to change it if 
> your system isn't like he one we envisaged when that was written.
> 
>> so, given that cms.perf was not set up should i understand that the 
>> specifications of cpu and io in cms.sched are just ignored as there is 
>> no facility for load reporting?
> Yes, the values were always set to zero.
uh oh .. i think that on cms.sched documentation should be a BOLD note, 
specifying that the option is invalid without cms.perf

additionally, i think that cms.perf should default to use the script 
that is packaged with the xrootd to have a minimal functional tool

Thank you!
Adrian


########################################################################
Use REPLY-ALL to reply to list

To unsubscribe from the XROOTD-L list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-L&A=1