Hello Gerri
I just finished with the test you suggested in the last post.
In general, If I deactivate the Cache I get worse results by a factor ~2 in time. In ~cosuna/public/Proof I put some plots that show the timing (all of them reading input data from a dcache server)
Timing_local_dcache_cache30MB.png -> running local with cache of 30MB.
Timing_Proof1Worker_dcache_cache30MB.png -> running on proof with 1 worker and cache of 30 MB.
Timing_local_dcache_nocache.png -> running local without the cache
Timing_Proof1Worker_dcache_nocache.png -> running on proof with 1 worker and without the cache.
I also generated the TTreePerfStats files
ioperf_dcache_cache30MB.root -> running local reading from dcache with a Cache of 30MB.
ioperf_dcache_nocache.root -> running local reading from dcache without Cache.
ioperf_localdisk_cache30MB.root -> running local reading from localdisk with a Cache of 30MB.
They tell me that the Cache is working well (running a local job) but unfortunately, I dont know how to produce the same TTreePerfStats info in proof (Using TTreePerfStats).
So how can I check if the TreeCache is working in proof?
anyhow deactivating the Cache when running locally the job may take 200 seconds, while sending the job to proof with 1 worker takes ~1200 seconds. So I still dont understand this behaviour.
The input chain is a list of files (some of them are really small). I put an example of a root input file on the same public path:
mc09_7TeV.107705_AlpgenJimmyWtaunuNp5_pt20_e511_s703_s730_r1097_r1113.root
thanks, Carlos