[
https://issues.apache.org/jira/browse/YARN-972?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13725528#comment-13725528
]
Allen Wittenauer commented on YARN-972:
---------------------------------------
bq. Nodes in probably the majority of clusters are configured with more slots
than cores. This is sensible because many types of task do a lot of IO and do
not even saturate half of a single core.
I disagree. It isn't a sensible thing to do at all unless it *also* schedules
based upon IO characteristics in addition to processor needs. The system
eventually ends up in a death spiral:
P1: "We need more processes on this machine because the load isn't high!"
P2: "OK! I've put more of our IO intensive processes on this machine!"
P1: "Weird! The CPUs are now spending more time in IO wait! Let's add more
processes since we have more CPU to get it higher!"
...
I posit that the reason why (at least in Hadoop 1.x systems) there are more
tasks per cores is simple: the jobs are crap. They are spending more time
launching JVMs and getting scheduled than they are actually executing code. It
gives the illusion that Hadoop isn't scheduling efficiently. Unless one
recognizes that there is a tipping point in parallelism, most users are going
to keep increasing it in blind faith that "more tasks = faster always".
Also, yes, I want YARN-796, but I don't think that's an orthogonal discussion.
My opinion is that they are different facets of the same discussion: how do we
properly schedule in a mixed load environment. It's very hard to get it 100%
efficient for all cases. Some folks are going to have to suffer. If I had to
pick, let it be the folks with workloads that are either terribly written or
sleep a lot and don't require a lot of processor when they do wake up.
> Allow requests and scheduling for fractional virtual cores
> ----------------------------------------------------------
>
> Key: YARN-972
> URL: https://issues.apache.org/jira/browse/YARN-972
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: api, scheduler
> Affects Versions: 2.0.5-alpha
> Reporter: Sandy Ryza
> Assignee: Sandy Ryza
>
> As this idea sparked a fair amount of discussion on YARN-2, I'd like to go
> deeper into the reasoning.
> Currently the virtual core abstraction hides two orthogonal goals. The first
> is that a cluster might have heterogeneous hardware and that the processing
> power of different makes of cores can vary wildly. The second is that a
> different (combinations of) workloads can require different levels of
> granularity. E.g. one admin might want every task on their cluster to use at
> least a core, while another might want applications to be able to request
> quarters of cores. The former would configure a single vcore per core. The
> latter would configure four vcores per core.
> I don't think that the abstraction is a good way of handling the second goal.
> Having a virtual cores refer to different magnitudes of processing power on
> different clusters will make the difficult problem of deciding how many cores
> to request for a job even more confusing.
> Can we not handle this with dynamic oversubscription?
> Dynamic oversubscription, i.e. adjusting the number of cores offered by a
> machine based on measured CPU-consumption, should work as a complement to
> fine-granularity scheduling. Dynamic oversubscription is never going to be
> perfect, as the amount of CPU a process consumes can vary widely over its
> lifetime. A task that first loads a bunch of data over the network and then
> performs complex computations on it will suffer if additional CPU-heavy tasks
> are scheduled on the same node because its initial CPU-utilization was low.
> To guard against this, we will need to be conservative with how we
> dynamically oversubscribe. If a user wants to explicitly hint to the
> scheduler that their task will not use much CPU, the scheduler should be able
> to take this into account.
> On YARN-2, there are concerns that including floating point arithmetic in the
> scheduler will slow it down. I question this assumption, and it is perhaps
> worth debating, but I think we can sidestep the issue by multiplying
> CPU-quantities inside the scheduler by a decently sized number like 1000 and
> keep doing the computations on integers.
> The relevant APIs are marked as evolving, so there's no need for the change
> to delay 2.1.0-beta.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
