[
https://issues.apache.org/jira/browse/KYLIN-1636?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261998#comment-15261998
]
liyang commented on KYLIN-1636:
-------------------------------
By MemMappedFileScanBenchmark
10000000 records read without mmap, 5485K rec/sec
10000000 records read with mmap, 5614K rec/sec
50 bytes per record, so the IO speed is ~250MB/sec which is far beyond the
~75MB/sec of normal mag disk. This is because the file was still in OS cache as
it's newly created. If we cap disk speed at 75MB/sec, then what to expect is
5500/(250/75) = *1650K* rec/sec
> Benchmark scan speed on different storage
> -----------------------------------------
>
> Key: KYLIN-1636
> URL: https://issues.apache.org/jira/browse/KYLIN-1636
> Project: Kylin
> Issue Type: Improvement
> Reporter: liyang
>
> We want to better understand the scan speed of different storage. Namely
> HBase, Cassandra, Parquet, MemMappedFile etc.
> To be specific, we test reading 10 millions records with single thread. Each
> record is 5 dimensions of type int4, and 2 measures of type long8, totally
> ~50 bytes if include cuboid ID and other overhead.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
