Here comes our contribution to discussion of Information modell and BES 
usecases.
The fact is that - as I mentioned in the meeting - we use interface part 
of the current BES spec. So we implemented a some pure web services 
which talks this BES interfaces.

The architecture is the following:
We have a lot of Linux cluster system in our country and we try to 
provide middleware to our users to use this resources as easy as it 
can.  Every cluster runs own  batch system like Condor or SGE and we 
choosed the BES as a common interface to this batch systems. So we have 
a services - currently it is called Job Controller (JC) but the name 
dosen\'t really important - which talks the BES and local batch system 
running on resources. The jobs are represented by the JSDL documents in 
our system so our scheduler service only make some decision which BES 
service will be used to run a jobs. It is true that in our system the JC 
which talks the BES interfaces does a lot of thing from the stage in and 
out files and translate JSDL to submit language of local batch system 
and monitor jobs. So in this case we have to describe resources with 
attributes to make decision on the scheduler. Currently, we are in the 
beginning of this work but actually we use these type of attributes to 
descriobe our cluster resources: We know this is the really toy 
description but it was useful in every day work:

LRMSType: type of Local Resource Manager services like Condor, PBS, LFS etc.
OperatingSystem: name of operating system of front end node of cluster 
(we assume here the homogenous cluster resources)
Architecture
CpuNum
RunningJob: jobs in running state in the queue
FreeCpuNum
OwnedCpuNum: how many cpu are used by the running jobs
JobNum: total length of queue
DataStorageTotal: total size of local disk space (here assume  shared  
storage system used by  cluster nodes)
DataStorageFree:
TempStorageTotal: total size of temporaly storage used by jobs (like 
size of  /tmp mount point)
TempStorageFree

We also use the BES interfaces in the SuperScheduler level of architecture.
We also have service which talks BES interface but it is represnets only 
a single desktop machine. In this case we are using the following 
attributes (the status is the same: this is only the toy but is is works):

OperatingSystem
Architecture
CpuNum: because it can be the SMP system
TotalPhysicalMemory
FreePhysicalMemory
TotalVirtualMemory
FreeVirtualMemory
LoadAvg: 5min average
DataStorageTotal
DataStorageFree
TempStorageTotal
TempStorageFree

We know these attributes and modell is less useful in general case but 
probably can be the a good starting point to understand what real grid 
system providers would like to see in resource descriptions and 
information modells.

-- 

Regards,
Ferenc Szalai
NIIF, Hungary
www.niif.hu