query caching /data caching

Navin_C

Hello All,

How does Vertica handle query caching and data caching.

I could not find any documents on this. 

Can we ask vertica to implement anyone of this explicitly.

I know it depends a lot on Linux File system caching, but how is it related.
We have 256 GB RAM on each node and the utilization would be 4-5%

Thanks

Abhishek_Rana

Hi,

Vertica does not cache queries (either plans or results). There is a modest cache of uncompressed data blocks Vertica maintains internally to avoid decompression on blocks reused quickly.
To invalidate this cache, run a Vertica query on lots of data unrelated to the query to measure (this may also help invalidate the kernel disk cache).

Vertica uses the Linux file-system cache for caching large amounts of data.The most likely caching effect is the kernel's use of memory to store recently accessed disk blocks.

On NUMA (Non-uniform memory access) machines, cached memory usage has been found low ir-respective of cache availability. The kernel default on NUMA machines with zone_reclaim_mode=1 does not play well with caches. Changing the configuration to 0 fixes the bad caching behavior. 

Regards'

Abhishek

Navin_C

what is the way to tell vertica to use maximum of memory(RAM)

There is a modest cache of uncompressed data blocks Vertica maintains internally to avoid decompression on blocks reused quickly.

Does this mean, Once I tried a select query and the result set gets stored in memory , then If I again fire the same query, vertica will access the data from memory .

And how do understand if the data is uncompressed while reading.