[
https://issues.apache.org/jira/browse/SPARK-22683?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16295770#comment-16295770
]
Xuefu Zhang commented on SPARK-22683:
-------------------------------------
I did some tests with the proposed change here and saw general efficiency
improvement with increasing tasksPerSlot. A number between 2 and 4 usually
gives the best trade off between performance and efficiency because the
efficiency gain is big while the performance impact is small. As the number
gets larger, the efficiency return diminishes while latency degradation becomes
unreasonable.
The effect of resource saving here plays best when the tasks are small. In such
a case, task finishes quickly, which doesn't require each task run on different
core. Because the tasks are small, the latency impact is also limited. On the
other hand, if the tasks are long running, keeping them waiting will increase
latency while resource saving is limited.
Based on what I observed, this knob, whatever is called, is practically useful
for efficiency conscious users. It doesn't seem hurting efficiency in any way.
For users that care about latency or don't care about cost, the default value
will do. Thus, I favor the inclusion of such a knob.
Nevertheless, I do agree that this knob is a little bit confusion as it is
because the concept of execution slot doesn't exist in Spark. I think it might
be better to define as a factor of the number of executors to be requested vs
the number of executors required to run all waiting tasks. So, a name like
{{spark.executor.conservation.factor}} might do. And it doesn't have to be an
integer.
With this included, I think the next thing to look at is the exponential
ramp-up in scheduling. I'd think up-front full allocation, in combination of
the conservation factor here, will have a big impact on resource efficiency.
> DynamicAllocation wastes resources by allocating containers that will barely
> be used
> ------------------------------------------------------------------------------------
>
> Key: SPARK-22683
> URL: https://issues.apache.org/jira/browse/SPARK-22683
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 2.1.0, 2.2.0
> Reporter: Julien Cuquemelle
> Labels: pull-request-available
>
> While migrating a series of jobs from MR to Spark using dynamicAllocation,
> I've noticed almost a doubling (+114% exactly) of resource consumption of
> Spark w.r.t MR, for a wall clock time gain of 43%
> About the context:
> - resource usage stands for vcore-hours allocation for the whole job, as seen
> by YARN
> - I'm talking about a series of jobs because we provide our users with a way
> to define experiments (via UI / DSL) that automatically get translated to
> Spark / MR jobs and submitted on the cluster
> - we submit around 500 of such jobs each day
> - these jobs are usually one shot, and the amount of processing can vary a
> lot between jobs, and as such finding an efficient number of executors for
> each job is difficult to get right, which is the reason I took the path of
> dynamic allocation.
> - Some of the tests have been scheduled on an idle queue, some on a full
> queue.
> - experiments have been conducted with spark.executor-cores = 5 and 10, only
> results for 5 cores have been reported because efficiency was overall better
> than with 10 cores
> - the figures I give are averaged over a representative sample of those jobs
> (about 600 jobs) ranging from tens to thousands splits in the data
> partitioning and between 400 to 9000 seconds of wall clock time.
> - executor idle timeout is set to 30s;
>
> Definition:
> - let's say an executor has spark.executor.cores / spark.task.cpus taskSlots,
> which represent the max number of tasks an executor will process in parallel.
> - the current behaviour of the dynamic allocation is to allocate enough
> containers to have one taskSlot per task, which minimizes latency, but wastes
> resources when tasks are small regarding executor allocation and idling
> overhead.
> The results using the proposal (described below) over the job sample (600
> jobs):
> - by using 2 tasks per taskSlot, we get a 5% (against -114%) reduction in
> resource usage, for a 37% (against 43%) reduction in wall clock time for
> Spark w.r.t MR
> - by trying to minimize the average resource consumption, I ended up with 6
> tasks per core, with a 30% resource usage reduction, for a similar wall clock
> time w.r.t. MR
> What did I try to solve the issue with existing parameters (summing up a few
> points mentioned in the comments) ?
> - change dynamicAllocation.maxExecutors: this would need to be adapted for
> each job (tens to thousands splits can occur), and essentially remove the
> interest of using the dynamic allocation.
> - use dynamicAllocation.backlogTimeout:
> - setting this parameter right to avoid creating unused executors is very
> dependant on wall clock time. One basically needs to solve the exponential
> ramp up for the target time. So this is not an option for my use case where I
> don't want a per-job tuning.
> - I've still done a series of experiments, details in the comments.
> Result is that after manual tuning, the best I could get was a similar
> resource consumption at the expense of 20% more wall clock time, or a similar
> wall clock time at the expense of 60% more resource consumption than what I
> got using my proposal @ 6 tasks per slot (this value being optimized over a
> much larger range of jobs as already stated)
> - as mentioned in another comment, tampering with the exponential ramp up
> might yield task imbalance and such old executors could become contention
> points for other exes trying to remotely access blocks in the old exes (not
> witnessed in the jobs I'm talking about, but we did see this behavior in
> other jobs)
> Proposal:
> Simply add a tasksPerExecutorSlot parameter, which makes it possible to
> specify how many tasks a single taskSlot should ideally execute to mitigate
> the overhead of executor allocation.
> PR: https://github.com/apache/spark/pull/19881
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
