Hello, I have a problem with a task of postgres i am running. The task reads 6.5GBs of data and if it has 7GB of memory it performs in about 20-30 seconds, else it takes more than 40-50 seconds to finish. This was not measured on Mesos btw.
The problem is that on Mesos even if i am assigning to task 1GB of memory, or 30GBs the execution time remains the same. I have enabled cgroups/mem In the same container i perform two runs. Before the first run i drop_caches and the execution time no matter the amount of memory is 40 seconds. After i perform the second run. The execution time no matter the amount of memory remains the same at 20-30 seconds i used `EXPLAIN` on the query, and no matter the resources given the hits are the same. On the first run it makes about 100 for example and on the second about 1000. I can't determine what the problem is, and i have to figure it out. Maybe there is no isolation for buffer I/O cache? I hope that someone more experienced than me can notice the problem. Its like it caches all the data. I have also tried assigning 64MBs of memory. The hits in cache remains the same, and so the execution time on both runs. Maybe i should enable another isolator of cgroups? Thank you, Zois Theodoros ---------------------- Institute of Computer Science FORTH-ICS Hellas
