[
https://issues.apache.org/jira/browse/HBASE-13259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14509896#comment-14509896
]
Zee Chen commented on HBASE-13259:
----------------------------------
[~ndimiduk] I have run the comparison between mmap and fileio and mmap and
offheap quite a few times. My fuzzy memory of observations from 2 months ago
were:
- mmap is identical to offheap in both latency and throughput
- mmap is slightly better than file in latency but it has greater throughput
Since offheap engine also allocates memory using mmap(), except using anonymous
maps as opposed to file backed map, the identical performance result is not
particularly surprising. In both cases the bulk of the cost is in copying the
data from mmap areas and we avoid a trip to the kernel.
I haven't tested this with those cache options you mentioned yet. I did test
with a few different block sizes and bucket sizes, but haven't found any
significant impact on latency yet.
> mmap() based BucketCache IOEngine
> ---------------------------------
>
> Key: HBASE-13259
> URL: https://issues.apache.org/jira/browse/HBASE-13259
> Project: HBase
> Issue Type: New Feature
> Components: BlockCache
> Affects Versions: 0.98.10
> Reporter: Zee Chen
> Fix For: 2.0.0, 1.2.0
>
> Attachments: HBASE-13259-v2.patch, HBASE-13259.patch, ioread-1.svg,
> mmap-0.98-v1.patch, mmap-1.svg, mmap-trunk-v1.patch
>
>
> Of the existing BucketCache IOEngines, FileIOEngine uses pread() to copy data
> from kernel space to user space. This is a good choice when the total working
> set size is much bigger than the available RAM and the latency is dominated
> by IO access. However, when the entire working set is small enough to fit in
> the RAM, using mmap() (and subsequent memcpy()) to move data from kernel
> space to user space is faster. I have run some short keyval gets tests and
> the results indicate a reduction of 2%-7% of kernel CPU on my system,
> depending on the load. On the gets, the latency histograms from mmap() are
> identical to those from pread(), but peak throughput is close to 40% higher.
> This patch modifies ByteByfferArray to allow it to specify a backing file.
> Example for using this feature: set hbase.bucketcache.ioengine to
> mmap:/dev/shm/bucketcache.0 in hbase-site.xml.
> Attached perf measured CPU usage breakdown in flames graph.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
