Disable caching


#1

I am executing some queries to compare some different chunks configurations I am using for my data. And I’d like to disable any caching, so SciDB will need to retrieve data directly from the disk. Is there some way to empty the cache between the execution of 2 consecutive queries? Or even to disable caching at all? Starting and stoping the services everytime is just too slow.

I’ve alrealdy set: smgr-cache-size=0 and stopped and restarted my cluster, but if I execute the same query twice, the first execution is considerably slower than the second one. So I assume that SciDB is caching the chunks.

Chaging the config.ini file and restarting the cluster is enough to apply the configuration? Is there another way to do this?

Thanks in advance


#2

Hi,

[quote]I’ve alrealdy set: smgr-cache-size=0 and stopped and restarted my cluster, but if I execute the same query twice, the first execution is considerably slower than the second one. So I assume that SciDB is caching the chunks.
[/quote]

I think part of the secret here is that the operating system itself has caching. That is what you are seeing here. Refer to linux documentation on how to flush and tune the filesystem cache. If you have any results, we’d be interested to see them!


#3

This link might be useful: unix.stackexchange.com/questions … nux-system


#4

Thank you a lot.

I cleaned up the OS cache for all the nodes, and the query execution time remais the same for consecutive executions now.