Moving over to the right mailing list.
On Sun, Mar 30, 2014 at 2:26 PM, WeiZheng Zhou <[email protected]>wrote: > Hi, > > We are doing some experiments requiring us to measure the latency > correctly. However, the latency varies a lot from run to run. As shown in > the below figure, we generated five tables, namely table 1 to table 5. > These tables have exactly the same schema, i.e. the same number of rows, > the same number of columns etc. Then we run the following code on these > tables: > > for each run from 1:10 > for each table 1:5 > connect to phoenix > select * from table where value > *9000* //max value =10,000, > therefore select 10% of total rows > measure the latency > close connection > endfor > endfor > > for each run from 1:10 > for each table 1:5 > connect to phoenix > select * from table where value > *5000* //max value =10,000, > therefore select 50% of total rows > measure the latency > close connection > endfor > endfor > > Since, these tables have the same schema and we run the same queries. We > should expect the same latency. > > In the first figure, we select value > 9000. As we can see, the latency > on each tables agrees well after the second run. However, at the first run, > the latency of table 1 is much longer than other tables. So, what makes the > table 1 differ from other tables, considering they have the same contents > and run the same query ? > > In the second figure, we select value > 5000 > Maybe because of cache, the latency of table 1 is decreased after the > first run. However, for other tables, the latencies are increasing, not by > a small amount by almost doubled. Do you have any idea about this large > increasing ? > > So, my question is, what are the possible features of phoenix, hbase can > explain this increasing and decreasing trend of latency ? > > Thanks a lot > > Weizheng Zhou > > > -- > You received this message because you are subscribed to the Google Groups > "Phoenix HBase User" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- Dan Di Spaltro
