[
https://issues.apache.org/jira/browse/YARN-2681?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
cntic updated YARN-2681:
------------------------
Attachment: yarn-site.xml
1) Configuration for testing HDFS Bandwidth Enforcement (yarn-site,xml)
- Enable enforcement: yarn.nodemanager.hdfs-bandwidth-enforcement.enable = true
- Port which Datanodes are listening:
yarn.nodemanager.hdfs-bandwidth-enforcement.port = 50010
- Devices' list of machine of data node:
yarn.nodemanager.hdfs-bandwidth-enforcement.devices = lo, eth0
- Interval for checking new tc config from persistence (ms):
yarn.nodemanager.hdfs-bandwidth-enforcement.check-tc-config-interval = 1000
- Since only the API for Resource has been upgrated to get/set HDFS Bandwidth
Enforcement, but the ResourceRequest has not been implemented yet, for test
purpose, the rate and burst use to define tc class can be given in YARN
configuration file:
+ test rate will be writen to an HDFS file:
yarn.nodemanager.hdfs-bandwidth-enforcement.test-rate-file = test-rate-file
+ test rate file example content : 30mbps ( rate unit: kbps, mbps, kbit,
mbit. See also: http://lartc.org/manpages/tc.txt)
+ test burst : yarn.nodemanager.hdfs-bandwidth-enforcement.test-burst = 50Kb
2) The patch is tested by running TestDFSIO-read
> Support bandwidth enforcement for containers while reading from HDFS
> --------------------------------------------------------------------
>
> Key: YARN-2681
> URL: https://issues.apache.org/jira/browse/YARN-2681
> Project: Hadoop YARN
> Issue Type: New Feature
> Components: capacityscheduler, nodemanager, resourcemanager
> Affects Versions: 2.5.1
> Environment: Linux
> Reporter: cntic
> Attachments: HADOOP-2681.patch, HADOOP-2681.patch, Traffic Control
> Design.png, yarn-site.xml
>
>
> To read/write data from HDFS on data node, applications establise TCP/IP
> connections with the datanode. The HDFS read can be controled by setting
> Linux Traffic Control (TC) subsystem on the data node to make filters on
> appropriate connections.
> The current cgroups net_cls concept can not be applied on the node where the
> container is launched, netheir on data node since:
> - TC hanldes outgoing bandwidth only, so it can be set on container node
> (HDFS read = incoming data for the container)
> - Since HDFS data node is handled by only one process, it is not possible
> to use net_cls to separate connections from different containers to the
> datanode.
> Tasks:
> 1) Extend Resource model to define bandwidth enforcement rate
> 2) Monitor TCP/IP connection estabilised by container handling process and
> its child processes
> 3) Set Linux Traffic Control rules on data node base on address:port pairs in
> order to enforce bandwidth of outgoing data
> Concept:
> http://www.hit.bme.hu/~do/papers/EnforcementDesign.pdf
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
