You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Craig Ingram (JIRA)" <ji...@apache.org> on 2017/12/07 13:32:01 UTC
[jira] [Commented] (YARN-7327) CapacityScheduler: Allocate
containers asynchronously by default
[ https://issues.apache.org/jira/browse/YARN-7327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16281840#comment-16281840 ]
Craig Ingram commented on YARN-7327:
------------------------------------
I finally got around to trying out asynchronous container allocation in Hadoop 2.9 and 3.0-SNAPSHOT (built from master a few days ago) with Spark 2.3-SNAPSHOT (built same day as Hadoop). This is all running on the same hardware described above (I did not repeat the tests on VMs). The test results are attached as is the jupyter notebook I used to create it. I did change the test from what was done above slightly by tweaking the core counts requested each round. It's now requesting 16, 32, 64, 128, and 256 whereas it was requesting 2, 20, 50, and 100 before. I reran the 2.7.3 tests as well. I also ran the 2.9 test with 4 threads and it came out basically the same as the 3.0 test with 4 threads; therefore I did not include it in the graphs.
||Legend||Test||
|sync3|synchronous 3.0-SNAPSHOT|
|sync29|synchronous 2.9|
|sync273|synchronous 2.7.3|
|async1-3|async with 1 thread on 3.0-SNAPSHOT|
|async1-29|async with 1 thread on 2.9|
|async1-273|async with 1 thread on 2.7.3|
|async2-3|async with 2 threads on 3.0-SNAPSHOT|
|async4-3|async with 4 threads on 3.0-SNAPSHOT|
|async8-3|async with 8 threads on 3.0-SNAPSHOT|
|async16-3|async with 16 threads on 3.0-SNAPSHOT|
[^async-scheduling-results.md]
[^schedule-async.png]
[^spark-on-yarn-schedule-async.ipynb]
While the numbers aren't as great as I was hoping (especially at higher thread pool counts), it's still a big improvement. I was mainly surprised by the flattening out of containers allocations per second at higher container counts. I was thinking of giving the RM more memory or at least looking into whether it is under GC pressure. Is there anywhere else I should look to tune this? Thanks!
> CapacityScheduler: Allocate 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
> Attachments: async-scheduling-results.md, schedule-async.png, spark-on-yarn-schedule-async.ipynb, yarn-async-scheduling.png
>
>
> 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: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org