[
https://issues.apache.org/jira/browse/BEAM-13082?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17431413#comment-17431413
]
Robert Burke commented on BEAM-13082:
-------------------------------------
Running the benchmark before and after 5 times, we can see the improvements.
{{benchstat old.txt new.txt
name old time/op new time/op delta
DataWriter/4B-8 11.1ns ±25% 5.5ns ±10% -50.16% (p=0.008 n=5+5)
DataWriter/16B-8 22.2ns ±15% 6.0ns ± 4% -72.87% (p=0.008 n=5+5)
DataWriter/1KB-8 949ns ±10% 51ns ±22% -94.64% (p=0.008 n=5+5)
DataWriter/4KB-8 3.47µs ± 3% 0.16µs ±30% -95.47% (p=0.008 n=5+5)
DataWriter/100KB-8 72.6µs ± 7% 4.3µs ±12% -94.01% (p=0.008 n=5+5)
DataWriter/1MB-8 313µs ± 5% 74µs ±21% -76.34% (p=0.008 n=5+5)
DataWriter/10MB-8 1.28ms ± 3% 1.15ms ±10% ~ (p=0.056 n=5+5)
DataWriter/100MB-8 15.6ms ±18% 12.4ms ± 2% -20.91% (p=0.008 n=5+5)
DataWriter/256MB-8 55.4ms ±87% 31.8ms ± 2% -42.48% (p=0.008 n=5+5)
name old alloc/op new alloc/op delta
DataWriter/4B-8 23.0B ± 0% 0.0B ~ (p=0.079 n=4+5)
DataWriter/16B-8 91.0B ± 0% 0.0B ~ (p=0.079 n=4+5)
DataWriter/1KB-8 5.89kB ± 0% 0.00kB ±100% -99.99% (p=0.008 n=5+5)
DataWriter/4KB-8 23.6kB ± 0% 0.0kB ±25% -99.99% (p=0.008 n=5+5)
DataWriter/100KB-8 492kB ± 0% 0kB ±14% -99.99% (p=0.008 n=5+5)
DataWriter/1MB-8 2.27MB ± 0% 0.00MB ±22% -99.98% (p=0.008 n=5+5)
DataWriter/10MB-8 10.5MB ± 0% 0.0MB ±13% -99.90% (p=0.008 n=5+5)
DataWriter/100MB-8 105MB ± 0% 1MB ± 7% -98.93% (p=0.008 n=5+5)
DataWriter/256MB-8 268MB ± 0% 8MB ± 6% -97.14% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
DataWriter/4B-8 0.00 0.00 ~ (all equal)
DataWriter/16B-8 0.00 0.00 ~ (all equal)
DataWriter/1KB-8 0.00 0.00 ~ (all equal)
DataWriter/4KB-8 0.00 0.00 ~ (all equal)
DataWriter/100KB-8 0.00 0.00 ~ (all equal)
DataWriter/1MB-8 2.00 ± 0% 1.00 ± 0% -50.00% (p=0.008 n=5+5)
DataWriter/10MB-8 6.00 ± 0% 5.00 ± 0% -16.67% (p=0.008 n=5+5)
DataWriter/100MB-8 6.00 ± 0% 5.00 ± 0% -16.67% (p=0.008 n=5+5)
DataWriter/256MB-8 6.00 ± 0% 5.00 ± 0% -16.67% (p=0.008 n=5+5)}}
> [Go SDK] Reduce churn in dataWriter by retaining byte slice.
> ------------------------------------------------------------
>
> Key: BEAM-13082
> URL: https://issues.apache.org/jira/browse/BEAM-13082
> Project: Beam
> Issue Type: Improvement
> Components: sdk-go
> Reporter: Robert Burke
> Assignee: Robert Burke
> Priority: P2
>
> It's been noted that we can reduce allocations and GC overhead produced by
> the dataWriter if we change the `w.buf = nil` to `w.buf = w.buf[:0]`. We
> should still nil out the buffer after the final flush in Close() however, to
> avoid retaining larger byte buffers after bundle termination.
> A dataWriter is created per bundle, and is only used and is safe to use by
> that bundle 's processing thread. Further, GRPC's Send call doesn't maintain
> ownership of the Proto message data after Send returns, allowing this re-use.
> A later optimization could use a sync.Pool to maintain a "freelist" of
> buffers to further reduce per bundle allocations but this would likely only
> be noticeable in streaming contexts. Such a free list should have a cap of
> keeping buffers under some threshold (say slices under 64MB in cap) to avoid
> retaining overly large buffers that aren't in active use. This idea though is
> out of scope for a first pass.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
