[
https://issues.apache.org/jira/browse/THRIFT-511?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12714513#action_12714513
]
Bryan Duxbury commented on THRIFT-511:
--------------------------------------
I tested out this patch's performance difference. Before:
{code}
10
user system total real
string set create 0.000000 0.000000 0.000000 ( 0.000143)
struct set create 0.010000 0.000000 0.010000 ( 0.003211)
100
user system total real
string set create 0.000000 0.000000 0.000000 ( 0.001208)
struct set create 0.330000 0.010000 0.340000 ( 0.337008)
1000
user system total real
string set create 0.030000 0.000000 0.030000 ( 0.037109)
struct set create 33.420000 0.190000 33.610000 ( 34.300607)
{code}
after:
{code}
10
user system total real
string set create 0.000000 0.000000 0.000000 ( 0.000243)
struct set create 0.000000 0.000000 0.000000 ( 0.001049)
100
user system total real
string set create 0.010000 0.000000 0.010000 ( 0.001654)
struct set create 0.000000 0.000000 0.000000 ( 0.006906)
1000
user system total real
string set create 0.040000 0.000000 0.040000 ( 0.036426)
struct set create 0.080000 0.000000 0.080000 ( 0.090610)
{code}
The performance boost is pretty stark. Also, there's zero testing to be added,
because verifying that hash codes for == structs before would have worked
equally well. I think this patch is ready to go.
If no one objects, I will commit it later today.
> Better performing hash method for generated structs
> ---------------------------------------------------
>
> Key: THRIFT-511
> URL: https://issues.apache.org/jira/browse/THRIFT-511
> Project: Thrift
> Issue Type: New Feature
> Components: Library (Ruby)
> Reporter: Bryan Duxbury
> Assignee: Dan Kinder
> Fix For: 0.2
>
> Attachments: thrift-511.patch
>
>
> As discussed in THRIFT-231, ruby generated structs' hash code method just
> always returns zero. While this is semantically correct, it leads to hashes
> having O(n) performance instead of O(1), which is critical for some
> applications.
> We can either take the approach that the Java library uses (optionally
> produce complex hash function), or just generate a good hash all the time.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
