crunch-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Wills (JIRA)" <>
Subject [jira] [Commented] (CRUNCH-294) Cost-based job planning
Date Sat, 16 Nov 2013 17:19:22 GMT


Josh Wills commented on CRUNCH-294:

The thumbnails looked cool from my perspective-- thanks!

IIRC, the original issue was concerned w/the fact that S2 and S3 were computationally expensive
operations, and should only be run once, which is the user was annoyed that the planner (which
is focused on minimizing disk IO) was running them twice: once in the reducer of one job,
and then again in the mapper of the second job-- she was less concerned w/disk IO, and more
concerned with overall throughput. So perhaps the issue is that we don't have a concept of
a CPU-intensive or computationally intensive DoFn-- DoFns can only signal (via scaleFactor)
their relative IO costs.

What about a new DoFn method that was like runAtMostOnce() which would ensure that a DoFn
was only ever run once, even if it cost more IO to do so? You could also argue that if you
had 2+ memory-intensive DoFns, you should try to run them in separate jobs so that their memory
usage wouldn't overwhelm the limits for the JVM, so that could be something else worth signaling
to the planner.

I think those are the major dimensions we care about, right? Disk IO primarily, then CPU/memory
usage? So we mainly want to optimize for Disk IO, except when there are one of these exceptional
conditions on a particular DoFn? Or is there a more elegant way to do this?

> Cost-based job planning
> -----------------------
>                 Key: CRUNCH-294
>                 URL:
>             Project: Crunch
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Josh Wills
>            Assignee: Josh Wills
>         Attachments: CRUNCH-294.patch, jobplan-default-new.png, jobplan-default-old.png,
jobplan-large_s2_s3.png, jobplan-lopsided.png
> A bug report on the user list drove me to revisit some of the core planning logic, particularly
around how we decide where to split up DoFns between two dependent MapReduce jobs.
> I found an old TODO about using the scale factor from a DoFn to decide where to split
up the nodes between dependent GBKs, so I implemented a new version of the split algorithm
that takes advantage of how we've propagated support for multiple outputs on both the map
and reduce sides of a job to do finer-grained splits that use information from the scaleFactor
calculations to make smarter split decisions.
> One high-level change along with this: I changed the default scaleFactor() value in DoFn
to 0.99f to slightly prefer writes that occur later in a pipeline flow by default.

This message was sent by Atlassian JIRA

View raw message