[
https://issues.apache.org/jira/browse/YARN-7327?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Craig Ingram updated YARN-7327:
-------------------------------
Description:
I was recently doing some research into Spark on YARN's startup time and
observed slow, synchronous allocation of containers/executors. I am testing on
a 4 node bare metal cluster w/48 cores and 128GB memory per node. YARN was only
allocating about 3 containers per second. Moreover when starting 3 Spark
applications at the same time with each requesting 44 containers, the first
application would get all 44 requested containers and then the next application
would start getting containers and so on.
>From looking at the code, it appears this is by design. There is an
>undocumented configuration variable that will enable asynchronous allocation
>of containers. I'm sure I'm missing something, but why is this not the
>default? Is there a bug or race condition in this code path? I've done some
>testing with it and it's been working and is significantly faster.
Here's the config:
`yarn.scheduler.capacity.schedule-asynchronously.enable`
Any help understanding this would be appreciated.
Thanks,
Craig
If you're curious about the performance difference with this setting, here are
the results:
The following tool was used for the benchmarks:
https://github.com/SparkTC/spark-bench
h2. async scheduler research
The goal of this test is to determine if running Spark on YARN with async
scheduling of containers reduces the amount of time required for an application
to receive all of its requested resources. This setting should also reduce the
overall runtime of short-lived applications/stages or notebook paragraphs. This
setting could prove crucial to achieving optimal performance when sharing
resources on a cluster with dynalloc enabled.
h3. Test Setup
Must update /etc/hadoop/conf/capacity-scheduler.xml (or through Ambari) between
runs.
`yarn.scheduler.capacity.schedule-asynchronously.enable=true|false`
conf files request executors counts of:
* 2
* 20
* 50
* 100
The apps are being submitted to the default queue on each cluster which caps at
48 cores on dynalloc and 72 cores on baremetal. The default queue was expanded
for the last two tests on baremetal so it could potentially take advantage of
all 144 cores.
h3. Test Environments
h4. dynalloc
4 VMs in Fyre (1 master, 3 workers)
8 CPUs/16 GB per node
model name : QEMU Virtual CPU version 2.5+
h4. baremetal
4 baremetal instances in Fyre (1 master, 3 workers)
48 CPUs/128GB per node
model name : Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
h3. Using spark-bench with timedsleep workload sync
h4. dynalloc
|| requested containers | avg | stdev||
|2 | 23.814900 | 1.110725|
|20 | 29.770250 | 0.830528|
|50 | 44.486600 | 0.593516|
|100 | 44.337700 | 0.490139|
h4. baremetal - 2 queues splitting cluster 72 cores each
|| requested containers | avg | stdev||
|2 | 14.827000 | 0.292290|
|20 | 19.613150 | 0.155421|
|50 | 30.768400 | 0.083400|
|100 | 40.931850 | 0.092160|
h4. baremetal - 1 queue to rule them all - 144 cores
|| requested containers | avg | stdev||
|2 | 14.833050 | 0.334061|
|20 | 19.575000 | 0.212836|
|50 | 30.765350 | 0.111035|
|100 | 41.763300 | 0.182700|
h3. Using spark-bench with timedsleep workload async
h4. dynalloc
|| requested containers | avg | stdev||
|2 | 22.575150 | 0.574296|
|20 | 26.904150 | 1.244602|
|50 | 44.721800 | 0.655388|
|100 | 44.570000 | 0.514540|
h5. 2nd run
|| requested containers | avg | stdev||
|2 | 22.441200 | 0.715875|
|20 | 26.683400 | 0.583762|
|50 | 44.227250 | 0.512568|
|100 | 44.238750 | 0.329712|
h4. baremetal - 2 queues splitting cluster 72 cores each
|| requested containers | avg | stdev||
|2 | 12.902350 | 0.125505|
|20 | 13.830600 | 0.169598|
|50 | 16.738050 | 0.265091|
|100 | 40.654500 | 0.111417|
h4. baremetal - 1 queue to rule them all - 144 cores
|| requested containers | avg | stdev||
|2 | 12.987150 | 0.118169|
|20 | 13.837150 | 0.145871|
|50 | 16.816300 | 0.253437|
|100 | 23.113450 | 0.320744|
was:
I was recently doing some research into Spark on YARN's startup time and
observed slow, synchronous allocation of containers/executors. I am testing on
a 4 node bare metal cluster w/48 cores and 128GB memory per node. YARN was only
allocating about 3 containers per second. Moreover when starting 3 Spark
applications at the same time with each requesting 44 containers, the first
application would get all 44 requested containers and then the next application
would start getting containers and so on.
>From looking at the code, it appears this is by design. There is an
>undocumented configuration variable that will enable asynchronous allocation
>of containers. I'm sure I'm missing something, but why is this not the
>default? Is there a bug or race condition in this code path? I've done some
>testing with it and it's been working and is significantly faster.
Here's the config:
`yarn.scheduler.capacity.schedule-asynchronously.enable`
Any help understanding this would be appreciated.
Thanks,
Craig
If you're curious about the performance difference with this setting, here are
the results:
The following tool was used for the benchmarks:
https://github.com/SparkTC/spark-bench
h2. async scheduler research
The goal of this test is to determine if running Spark on YARN with async
scheduling of containers reduces the amount of time required for an application
to receive all of its requested resources. This setting should also reduce the
overall runtime of short-lived applications/stages or notebook paragraphs. This
setting could prove crucial to achieving optimal performance when sharing
resources on a cluster with dynalloc enabled.
h3. Test Setup
Must update /etc/hadoop/conf/capacity-scheduler.xml (or through Ambari) between
runs.
`yarn.scheduler.capacity.schedule-asynchronously.enable=true|false`
conf files request executors counts of:
* 2
* 20
* 50
* 100
The apps are being submitted to the default queue on each cluster which caps at
48 cores on dynalloc and 72 cores on baremetal. The default queue was expanded
for the last two tests on baremetal so it could potentially take advantage of
all 144 cores.
h3. Test Environments
h4. dynalloc
4 VMs in Fyre (1 master, 3 workers)
8 CPUs/16 GB per node
model name : QEMU Virtual CPU version 2.5+
h4. baremetal
4 baremetal instances in Fyre (1 master, 3 workers)
48 CPUs/128GB per node
model name : Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
h3. Using spark-bench with timedsleep workload sync
h4. dynalloc
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 23.814900 | 1.110725|
|spark-on-yarn-schedule-async1.time | 29.770250 | 0.830528|
|spark-on-yarn-schedule-async2.time | 44.486600 | 0.593516|
|spark-on-yarn-schedule-async3.time | 44.337700 | 0.490139|
h4. baremetal - 2 queues splitting cluster 72 cores each
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 14.827000 | 0.292290|
|spark-on-yarn-schedule-async1.time | 19.613150 | 0.155421|
|spark-on-yarn-schedule-async2.time | 30.768400 | 0.083400|
|spark-on-yarn-schedule-async3.time | 40.931850 | 0.092160|
h4. baremetal - 1 queue to rule them all - 144 cores
||conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 14.833050 | 0.334061|
|spark-on-yarn-schedule-async1.time | 19.575000 | 0.212836|
|spark-on-yarn-schedule-async2.time | 30.765350 | 0.111035|
|spark-on-yarn-schedule-async3.time | 41.763300 | 0.182700|
h3. Using spark-bench with timedsleep workload async
h4. dynalloc
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 22.575150 | 0.574296|
|spark-on-yarn-schedule-async1.time | 26.904150 | 1.244602|
|spark-on-yarn-schedule-async2.time | 44.721800 | 0.655388|
|spark-on-yarn-schedule-async3.time | 44.570000 | 0.514540|
h5. 2nd run
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 22.441200 | 0.715875|
|spark-on-yarn-schedule-async1.time | 26.683400 | 0.583762|
|spark-on-yarn-schedule-async2.time | 44.227250 | 0.512568|
|spark-on-yarn-schedule-async3.time | 44.238750 | 0.329712|
h4. baremetal - 2 queues splitting cluster 72 cores each
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 12.902350 | 0.125505|
|spark-on-yarn-schedule-async1.time | 13.830600 | 0.169598|
|spark-on-yarn-schedule-async2.time | 16.738050 | 0.265091|
|spark-on-yarn-schedule-async3.time | 40.654500 | 0.111417|
h4. baremetal - 1 queue to rule them all - 144 cores
|| conf | avg | stdev||
|spark-on-yarn-schedule-async0.time | 12.987150 | 0.118169|
|spark-on-yarn-schedule-async1.time | 13.837150 | 0.145871|
|spark-on-yarn-schedule-async2.time | 16.816300 | 0.253437|
|spark-on-yarn-schedule-async3.time | 23.113450 | 0.320744|
> Launch containers asynchronously by default
> -------------------------------------------
>
> Key: YARN-7327
> URL: https://issues.apache.org/jira/browse/YARN-7327
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Craig Ingram
> Priority: Trivial
>
> I was recently doing some research into Spark on YARN's startup time and
> observed slow, synchronous allocation of containers/executors. I am testing
> on a 4 node bare metal cluster w/48 cores and 128GB memory per node. YARN was
> only allocating about 3 containers per second. Moreover when starting 3 Spark
> applications at the same time with each requesting 44 containers, the first
> application would get all 44 requested containers and then the next
> application would start getting containers and so on.
>
> From looking at the code, it appears this is by design. There is an
> undocumented configuration variable that will enable asynchronous allocation
> of containers. I'm sure I'm missing something, but why is this not the
> default? Is there a bug or race condition in this code path? I've done some
> testing with it and it's been working and is significantly faster.
>
> Here's the config:
> `yarn.scheduler.capacity.schedule-asynchronously.enable`
>
> Any help understanding this would be appreciated.
>
> Thanks,
> Craig
>
> If you're curious about the performance difference with this setting, here
> are the results:
>
> The following tool was used for the benchmarks:
> https://github.com/SparkTC/spark-bench
> h2. async scheduler research
> The goal of this test is to determine if running Spark on YARN with async
> scheduling of containers reduces the amount of time required for an
> application to receive all of its requested resources. This setting should
> also reduce the overall runtime of short-lived applications/stages or
> notebook paragraphs. This setting could prove crucial to achieving optimal
> performance when sharing resources on a cluster with dynalloc enabled.
> h3. Test Setup
> Must update /etc/hadoop/conf/capacity-scheduler.xml (or through Ambari)
> between runs.
> `yarn.scheduler.capacity.schedule-asynchronously.enable=true|false`
> conf files request executors counts of:
> * 2
> * 20
> * 50
> * 100
> The apps are being submitted to the default queue on each cluster which caps
> at 48 cores on dynalloc and 72 cores on baremetal. The default queue was
> expanded for the last two tests on baremetal so it could potentially take
> advantage of all 144 cores.
> h3. Test Environments
> h4. dynalloc
> 4 VMs in Fyre (1 master, 3 workers)
> 8 CPUs/16 GB per node
> model name : QEMU Virtual CPU version 2.5+
> h4. baremetal
> 4 baremetal instances in Fyre (1 master, 3 workers)
> 48 CPUs/128GB per node
> model name : Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
> h3. Using spark-bench with timedsleep workload sync
> h4. dynalloc
> || requested containers | avg | stdev||
> |2 | 23.814900 | 1.110725|
> |20 | 29.770250 | 0.830528|
> |50 | 44.486600 | 0.593516|
> |100 | 44.337700 | 0.490139|
> h4. baremetal - 2 queues splitting cluster 72 cores each
> || requested containers | avg | stdev||
> |2 | 14.827000 | 0.292290|
> |20 | 19.613150 | 0.155421|
> |50 | 30.768400 | 0.083400|
> |100 | 40.931850 | 0.092160|
> h4. baremetal - 1 queue to rule them all - 144 cores
> || requested containers | avg | stdev||
> |2 | 14.833050 | 0.334061|
> |20 | 19.575000 | 0.212836|
> |50 | 30.765350 | 0.111035|
> |100 | 41.763300 | 0.182700|
> h3. Using spark-bench with timedsleep workload async
> h4. dynalloc
> || requested containers | avg | stdev||
> |2 | 22.575150 | 0.574296|
> |20 | 26.904150 | 1.244602|
> |50 | 44.721800 | 0.655388|
> |100 | 44.570000 | 0.514540|
> h5. 2nd run
> || requested containers | avg | stdev||
> |2 | 22.441200 | 0.715875|
> |20 | 26.683400 | 0.583762|
> |50 | 44.227250 | 0.512568|
> |100 | 44.238750 | 0.329712|
> h4. baremetal - 2 queues splitting cluster 72 cores each
> || requested containers | avg | stdev||
> |2 | 12.902350 | 0.125505|
> |20 | 13.830600 | 0.169598|
> |50 | 16.738050 | 0.265091|
> |100 | 40.654500 | 0.111417|
> h4. baremetal - 1 queue to rule them all - 144 cores
> || requested containers | avg | stdev||
> |2 | 12.987150 | 0.118169|
> |20 | 13.837150 | 0.145871|
> |50 | 16.816300 | 0.253437|
> |100 | 23.113450 | 0.320744|
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
