[UR-WG] SGAS extensions to usage record

[Henrik Thostrup Jensen]

Hi

Here is a list of the UR extensions used in SGAS. It is provided in the 
hope to be usefull in the UR2 creation process.

Arguably the most important extension the ability to describe VO 
affiliation, including groups and roles. It looks like this:

<vo:VO vo:type="voms" xmlns:vo="http://www.sgas.se/namespaces/2009/05/ur/vo">
     <vo:Name>bio.ndgf.org</vo:Name>
     <vo:Issuer>/O=Grid/O=NorduGrid/CN=host/voms.ndgf.org</vo:Issuer>
     <vo:Attribute>
         <vo:Group>bio.ndgf.org/test</vo:Group>
         <vo:Role>VO-dev</vo:Role>
     </vo:Attribute>
     <vo:Attribute>
         <vo:Group>bio.ndgf.org</vo:Group>
     </vo:Attribute>
     <vo:Attribute>
         <vo:Group>bio.ndgf.org/test</vo:Group>
     </vo:Attribute>
</vo:VO>

It is placed in the user identity block. It is somewhat closely modelled 
after VOMS (which is what we needed to model), but has been used for 
several other systems (usually one would just the Name and type elements). 
Descibing the type VO information a rather important feature IMO.

Futhermore we have a range of small simple extensions. We share submit 
time with the DEISA project (well we just reused what they had made), 
though there is no data exchange between NDGF and DEISA. The following 
lists the extensions:

Submit time (shared with Deisa)
<deisa:SubmitTime xmlns:deisa="http://rmis.deisa.org/acct">2009-07-07T09:06:37Z</deisa:SubmitTime>
Submission time can be used to tell how long time a job spend in queue, which is an often-asked-for metric.

SGAS UR extensions:

KSI2KCpuDuration (scaled cpu duration)
<sgas:KSI2KCpuDuration>PT150.4S</sgas:KSI2KCpuDuration>

KSI2KWallDuration (scaled wall duration)
<sgas:KSI2KWallDuration>PT8488.44S</sgas:KSI2KWallDuration>

The two ksi2k elements are no longer used. If there should be the 
possibility for scaled values in ur2, it should not be limited to a 
specific scaling method.

insertTime (when the record was inserted into the database. this field only
exists when extracting a record from a database
<ur:RecordIdentity sgas:insertTime="2009-07-07T09:06:52Z" />

UserTime (time spend in userland)
<sgas:UserTime>PT84.3S</sgas:UserTime>

KernelTime (time spend in kernel time)
<sgas:KernelTime>PT13.2S</sgas:KernelTime>

ExitCode (job exit code, integer value)
<sgas:ExitCode>0</sgas:ExitCode>

MajorPageFaults (times a major page fault occured)
<sgas:MajorPageFaults>3</sgas:MajorPageFaults>

RuntimeEnvironment (requested ARC runtime environments, allows 
per-application accounting, the element can be repeated)
<sgas:RuntimeEnvironment>APPS/HEP/ATLAS-13.0.40</sgas:RuntimeEnvironment>

The latter five where actually requested by CSC (Finish HPC centre), which 
also uses SGAS (there are at least five deployments of the SGAS accounting 
service AFAIK).

Futhermore we are thinking about adding fields for how many files the job 
down/up-loaded and how much data was transferred. Furthermore how many 
files was used from the cache (Input files, which existed locally and does 
not need to be downloaded. This might be a bit ARC specific, but the 
concept could still be usefull for others).

We also include information about what software produced the record, and 
optionally which version of the software.  This information does not 
describe the job in any way, but can be practical for identifying the 
source of the record and detect incompatabilities.

logger = namespace http://www.sgas.se/namespaces/2010/08/logger

LoggerName (name of the logger, can have optional version element)
<logger:LoggerName>ARC0-URLogger</logger:LoggerName>
<logger:LoggerName logger:version="0.8.2.2">ARC0-URLogger</logger:LoggerName>

Other issues:

The specification text for the machine name, host names, etc. could be 
clearer in the standard.

In the future a sample set of records for compliancy testing could be 
provided.


     Best regards, Henrik

  Software Developer, Henrik Thostrup Jensen <htj at ndgf.org>
  Nordic Data Grid Facility. WWW: www.ndgf.org