[
https://issues.apache.org/jira/browse/HADOOP-8468?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13291584#comment-13291584
]
Konstantin Shvachko commented on HADOOP-8468:
---------------------------------------------
Junping, I went over the design document. It is pretty comprehensive. A few
comments on the design.
# Conceptually you are extending current Network Topology by introducing a new
layer of leaf nodes. Current topology assumes that physical nodes are the
leaves of the hierarchy and you add virtual nodes that can reside on physical
nodes. I think this is a more logical way to look at the new topology, rather
than saying that you introduce the second layer (node groups) over the nodes,
as document does.
# The document should clarify how local storage is used by VMs on a physical
box. I think the assumption is that VMs never share storage resources.
Otherwise there could be a reporting problem. That is, if two VMs share a drive
and send two DF reports to the NameNode, then the drive will be counted twice,
which can cause problems. I'd recommend to update the pictures and add a
section talking about reporting of DNs' resources to NN to make this issue
explicitly covered in the design.
# For block replication there are 3 policies to consider:
#* block placement policy, when a new block is created
#* block replication policy, when under-replicated blocks are recovered
#* replica removal policy, when replicas are removed for over-replicated blocks
You covered the first two, and probably need to look into the third as well.
For the first two I'd be good to write down the entire modified policy rather
than just listing the differences.
_And make sure they converge to existing policies if virtual node layer is not
defined._
# For YARN I am not convinced you will need to run multiple VMs per node, if
not for the sake of generosity. It seems YARN should rely on NodeManager to
report resources and manage Containers of a node as a whole. Not sure how
multiple VMs on a node can help here.
For MRv1 on the contrary running multiple VMs per node can be useful for
modeling variable slots. In this case again the VMs should not share memory
otherwise repoting will go wrong.
> Umbrella of enhancements to support different failure and locality topologies
> -----------------------------------------------------------------------------
>
> Key: HADOOP-8468
> URL: https://issues.apache.org/jira/browse/HADOOP-8468
> Project: Hadoop Common
> Issue Type: Bug
> Components: ha, io
> Affects Versions: 1.0.0, 2.0.0-alpha
> Reporter: Junping Du
> Assignee: Junping Du
> Priority: Critical
> Attachments: HADOOP-8468-total-v3.patch, HADOOP-8468-total.patch,
> Proposal for enchanced failure and locality topologies.pdf
>
>
> The current hadoop network topology (described in some previous issues like:
> Hadoop-692) works well in classic three-tiers network when it comes out.
> However, it does not take into account other failure models or changes in the
> infrastructure that can affect network bandwidth efficiency like:
> virtualization.
> Virtualized platform has following genes that shouldn't been ignored by
> hadoop topology in scheduling tasks, placing replica, do balancing or
> fetching block for reading:
> 1. VMs on the same physical host are affected by the same hardware failure.
> In order to match the reliability of a physical deployment, replication of
> data across two virtual machines on the same host should be avoided.
> 2. The network between VMs on the same physical host has higher throughput
> and lower latency and does not consume any physical switch bandwidth.
> Thus, we propose to make hadoop network topology extend-able and introduce a
> new level in the hierarchical topology, a node group level, which maps well
> onto an infrastructure that is based on a virtualized environment.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
