Impala is ~50% idle
Jun 7, 2013 at 7:17 pm
Just to clarify, is there a way to push the parallelism, or is it more like
'you get what you get' (i.e. it uses what it can)?
: Hi Amiesh, cluster usage is going to be very dependent on the query. Let me explain. First, the number of machines involved in answering a query depend is the number of machines that store data that is relevant to the query (i.e. after pruning irrelevant partitions). Second, if a machine is involved in answering a query it is not guaranteed to use all available CPU cores because we currently do not multi thread all parts of a plan fragment. Improvements are in the works to judiciously adjust
: There is currently no recommended user-accessible mechanism for dictating the degree of parallelism at any level of Impala. Alex On Fri, Jun 7, 2013 at 12:17 PM, Amlesh Jayakumar wrote:
Re: Impala Query
When will impala supports group_concat?
Re: Hive Header line in Select query help?
impala avro support
Re: Data replication in impala
Integration of impala and sqoop, talend
newbie - best mechanism to stream writes and the immediate reads in impala
ORDER BY rand() broken?
Re: Updates to Data stored in Parquet
Re: Working of execution engine in impala
3 of 4
Jun 6, '13 at 1:20a
Jun 7, '13 at 8:00p
2 users in discussion
Alex Behm (2)
Amlesh Jayakumar (2)
Groups & Organizations
site design / logo © 2022 Grokbase