Github user andrewor14 commented on the pull request:
https://github.com/apache/spark/pull/7279#issuecomment-122373273
Hi all, I spent some time benchmarking the performance improvements of this
feature. I ran the following on a 4-node `m1.xlarge` cluster with 6 executors,
each with `4g` of memory. The numbers reported here are the median over 10 runs.
**Micro-benchmark.** Checkpoint a ~700MB input RDD
- normal checkpoint: 9337ms
- local checkpoint: 1599ms
- improvement: *82.9%*
**Macro-benchmark.** Run EM LDA on a ~235MB file on HDFS
- `--maxIterations` = 20
- `--checkpointInterval` = 2
- `--k` = 10
- normal checkpoint: 96078ms
- local checkpoint: 79592ms
- improvement: *17.2%*
The performance improvement in LDA is not as dramatic as the micro
benchmark. This is because the majority of time is spent on actual computation
rather than checkpointing.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
