[
https://issues.apache.org/jira/browse/HDFS-9806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15319050#comment-15319050
]
Zhe Zhang commented on HDFS-9806:
---------------------------------
Another thought is, maybe we can leverage caching policies and consistency
models from NFS? Fundamentally, each "small HDFS" is like an NFS client, and
the "big external store" is like the NFS server.
E.g. maybe we can use lease-based locking to prevent conflicting updates to the
same subtree.
bq. Initially, writes are not supported through HDFS (read-only). Refresh is an
important case,
Thanks for the clarification. This happens to be the most important use case in
our setup. I think a read-only "small HDFS" should be able to simplify the
design. A few additional questions:
# Should NN periodically refresh for the entire mounted subtree? Or fetch new
metadata and data on-demand? Or a mix of on-demand fetch and prefetching? E.g.
when application accesses file {{/data/log1.txt}} and it's a caches miss on
small HDFS, proactively fetch all files under {{/data/}} to small HDFS. If we
assume small HDFS has a significantly smaller capacity than external store,
refreshing the entire subtree seems too heavy (network bandwidth usage and
small HDFS capacity)?
# On the on-demand fetching path, the block will be transferred from external
store to the small HDFS DN first, and them from small HDFS DN to application.
This actually increases latency from 1 hop to 2 hops. It's tricky how to reduce
this.
> Allow HDFS block replicas to be provided by an external storage system
> ----------------------------------------------------------------------
>
> Key: HDFS-9806
> URL: https://issues.apache.org/jira/browse/HDFS-9806
> Project: Hadoop HDFS
> Issue Type: New Feature
> Reporter: Chris Douglas
> Attachments: HDFS-9806-design.001.pdf
>
>
> In addition to heterogeneous media, many applications work with heterogeneous
> storage systems. The guarantees and semantics provided by these systems are
> often similar, but not identical to those of
> [HDFS|https://hadoop.apache.org/docs/current/hadoop-project-dist/hadoop-common/filesystem/index.html].
> Any client accessing multiple storage systems is responsible for reasoning
> about each system independently, and must propagate/and renew credentials for
> each store.
> Remote stores could be mounted under HDFS. Block locations could be mapped to
> immutable file regions, opaque IDs, or other tokens that represent a
> consistent view of the data. While correctness for arbitrary operations
> requires careful coordination between stores, in practice we can provide
> workable semantics with weaker guarantees.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
