[torqueusers] Upgrade from 2.1.*

Joshua Bernstein jbernstein at penguincomputing.com
Tue Aug 17 17:30:00 MDT 2010



Garrick Staples wrote:
> On Tue, Aug 17, 2010 at 04:18:58PM -0700, Joshua Bernstein alleged:
>>
>> Ken Nielson wrote:
>>> On 08/17/2010 12:14 PM, Glen Beane wrote:
>>>> that would be good.  Although I'd really love to get rid of just
>>>> dumping the struct right to disk and have something that can be
>>>> extended without jumping through hoops to maintain compatibility.
>>>>
>>>>    
>>> Did I hear XML?
>>>
>>> serverdb is now an XML file in trunk. How about job files as well?
>> No. I'm not in favor of doing something like this. Imagine a third party 
>> application, say like PBSQuery having now to parse XML!
> 
> PBSQuery reads the job files? Why would it do that? 

No, it doesn't read the job files themselves, but it does query the 
server for information about jobs. I always figured then, in turn 
pbs_server would have to read through its own job files to return the 
results.

>> It might work well for the small systems you guys mess around on, but 
>> I'd really like us all to consider what happens when hundreds of 
>> thousands of jobs are present, as even with the current implementation 
>> things are uberslow sometimes.
> 
> Obviously one doesn't just throw XML out there in the wild polluting the place.
> One releases the associated tools and libraries to read/write the XML.
> 
> This data is only ever read when pbs_server starts. Read performance isn't
> important. We only care about write performance.

What if pbs_server can't keep all of the job information in memory? 
Write performance would seem to be limited to the performance of the 
file system.

-Josh


More information about the torqueusers mailing list