Re: Disk I/O

On Jan 30, 2015, at 10:48, Tim Burress wrote:

On 01/30/2015 06:15 PM, Martin Kersten wrote:

...

Which version and on what platform are you running your queries? A snippet of your query could be helpful.

The version is Oct2014-SP1 (I'll upgrade to the new one as soon as some of these queries finish!) on Fedora 20. The hardware is a 3.4GHz Xeon CPU with 32GB of RAM.

The queries are all pretty simple, but the table involved has perhaps 4x10^9 records. A representative query looks like this:

select iso2 as country,count(distinct source) as sources from sessions where timestamp>='2014-10-01' and timestamp<'2015-01-01' and

This probably triggers the creation of imprints, that also causes some I/O. Do you notice any major exec. time difference if you run your query multiple times?

("action"='R' or "action"='T') and medium='E' group by country order by sources desc;

Given the various 'where' conditions, Roberto's idea -- the need to store intermediate results -- sounds very plausible, particularly when there are other jobs claiming RAM on the system.

I'm re-running it now with most other user processes stopped. The mserver5 process is using about 77% of memory and nmon shows the disk with the database is pegged at 100% utilization. All writes at the moment, though the df command doesn't show any changes in used space, so I'm not sure where the written data is going!

Thanks!

Tim

_______________________________________________ users-list mailing list users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list