Mailing-List: contact derby-dev-help@db.apache.org; run by ezmlm
Precedence: bulk
Reply-To: <derby-dev@db.apache.org>
Message-ID: <1393048.1164742162397.JavaMail.jira@brutus>
Date: Tue, 28 Nov 2006 11:29:22 -0800 (PST)
From: "Knut Anders Hatlen (JIRA)" <jira@apache.org>
To: derby-dev@db.apache.org
Subject: [jira] Updated: (DERBY-2122) Optimize ContainerLock.isCompatible()
In-Reply-To: <17558296.1164740064057.JavaMail.jira@brutus>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

     [ http://issues.apache.org/jira/browse/DERBY-2122?page=all ]

Knut Anders Hatlen updated DERBY-2122:
--------------------------------------

    Attachment: derby-2122-v1.diff
                derby-2122-v1.stat

The attached patch removes the compatibility matrix from ContainerLock and instead maintains a bit mask containing the type ids of the compatible lock types. Compatibility checking between two lock types could therefore be performed by a bitwise and of the the bit mask and the type id.

Similar changes were made to RowLock to keep the two classes consistent.

I have run some experiments with this patch and the test client from DERBY-1961 (single-record selects with 16 concurrent clients on embedded Derby).

Experiment 1: 40 runs (20 with clean trunk, 20 with patch for ContainerLock only) consisting of 30 seconds warm-up and 60 seconds counting transactions. Result: Average throughput increased by 1.2% with the patch.

Experiment 2: Same as above, but counting transactions for 100 seconds instead of 60 seconds. Result: Average throughput increased by 1.7% with the patch.

Experiment 3: Same as above, but with the full patch (that is, both ContainerLock and RowLock changed). Result: Average throughput increased by 1.9% with the patch.

I don't have a fancy script that calculates how significant these numbers are (can do it on request, since I really need to refresh my statistics skills), but they at least give an indication that the patch has a small positive impact.

Reviews would be appreciated. Thanks.

> Optimize ContainerLock.isCompatible()
> -------------------------------------
>
>                 Key: DERBY-2122
>                 URL: http://issues.apache.org/jira/browse/DERBY-2122
>             Project: Derby
>          Issue Type: Improvement
>          Components: Store, Performance
>    Affects Versions: 10.2.1.6
>            Reporter: Knut Anders Hatlen
>         Assigned To: Knut Anders Hatlen
>            Priority: Trivial
>         Attachments: derby-2122-v1.diff, derby-2122-v1.stat
>
>
> The numbers at http://wiki.apache.org/db-derby/Derby1961MethodCalls indicate that ContainerLock.isCompatible() is called many times per transaction, at least when there are multiple concurrent clients operating on the same containers. Currently, it looks into a two-dimensional array to find out whether two locks are compatible. This could be implemented more efficiently, for instance by having a bit pattern in each ContainerLock object representing which lock types it is compatible with.

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira