[
https://issues.apache.org/jira/browse/OAK-1379?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nicholas Marchi updated OAK-1379:
---------------------------------
Description:
We were creating some data for a performance test and we noticed that our
script took significantly longer to run with mongomk than with crx2.
The scenario we are running looks like this:
1.) Create x users, x groups and x resources.
2.) Add set of n users to a group #n and then add an access control entry for
that group to resource #n. So group 1 will have only user 1 in it and will be
assigned to resource 1. Group 10 will have users 1 - 10 in it and will be
assigned to resource 10.
What we found was that it took significantly longer to perform this action with
mongomk than it did using crx2. From 1 - 500 users the transactions took about
the same amount of time. Beyond 500 users mongoMK began to take significantly
more time with each additional user, to the point that at 500 users the
transaction took around 5 seconds in both cases, but at 1000 users the
transaction with crx2 took around 10 seconds but with mongomk it took ~40
seconds.
For the record both tests were performed on indentical systems (CentOS6x64 with
16gb ram) running load15, using either the crx3mongo, or crx2 run modes. In the
case of mongo, the database is running on a remote server. Only AEM is running
on the test server.
The scenario wasn't designed as a performance test, we were just bulk creating
test data, I just happened to print out the time with each update and noticed
that they were very different between crx3mongo and crx2.
was:
We were creating some data for a performance test and we noticed that our
script took significantly longer to run with mongomk than with crx2.
The scenario we are running looks like this:
1.) Create x users, x groups and x resources.
2.) Add set of n users to a group #n and then add an access control entry for
that group to resource #n. So group 1 will have only user 1 in it and will be
assigned to resource 1. Group 10 will have users 1 - 10 in it and will be
assigned to resource 10.
What we found was that it took significantly longer to perform this action with
mongomk than it did using crx2.
For the record both tests were performed on indentical systems (CentOS6x64 with
16gb ram) running load15, using either the crx3mongo, or crx2 run modes. In the
case of mongo, the database is running on a remote server. Only AEM is running
on the test server.
The scenario wasn't designed as a performance test, we were just bulk creating
test data, I just happened to print out the time with each update and noticed
that they were very different between crx3mongo and crx2.
> Group membership and ACL assignment takes significantly more time with
> mongomk than with crx2
> ---------------------------------------------------------------------------------------------
>
> Key: OAK-1379
> URL: https://issues.apache.org/jira/browse/OAK-1379
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: mongomk
> Affects Versions: 0.15
> Environment: Centos 6x64
> Reporter: Nicholas Marchi
> Attachments: ACLBenchmark-1.0-SNAPSHOT.zip, load15mongovscrx2.jpg
>
>
> We were creating some data for a performance test and we noticed that our
> script took significantly longer to run with mongomk than with crx2.
> The scenario we are running looks like this:
> 1.) Create x users, x groups and x resources.
> 2.) Add set of n users to a group #n and then add an access control entry for
> that group to resource #n. So group 1 will have only user 1 in it and will be
> assigned to resource 1. Group 10 will have users 1 - 10 in it and will be
> assigned to resource 10.
> What we found was that it took significantly longer to perform this action
> with mongomk than it did using crx2. From 1 - 500 users the transactions took
> about the same amount of time. Beyond 500 users mongoMK began to take
> significantly more time with each additional user, to the point that at 500
> users the transaction took around 5 seconds in both cases, but at 1000 users
> the transaction with crx2 took around 10 seconds but with mongomk it took ~40
> seconds.
> For the record both tests were performed on indentical systems (CentOS6x64
> with 16gb ram) running load15, using either the crx3mongo, or crx2 run modes.
> In the case of mongo, the database is running on a remote server. Only AEM is
> running on the test server.
> The scenario wasn't designed as a performance test, we were just bulk
> creating test data, I just happened to print out the time with each update
> and noticed that they were very different between crx3mongo and crx2.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
