Re: Results of a JR Oracle test that we conducted

[Marcin Nowak]

Hi,




BR,
Marcin Nowak

Sriram Narayanan wrote:

Could you put some info about the structure of repository such as node number, max and average depth and width of repository tree, did you use version mechanism or references between nodes?

Hi list:

We ran a few tests on JR and Oracle and have posted our results here.
We welcome comments from the community.

JR and Database:

All times are in milliseconds


5    1    313    703    2688    537
5    20    156    828    23793    237
10    5    156    938    14313    286
10    20    171    719    49475    247
20    5    187    812    27029    270
20    10    156    797    49588    247
50    5    156    844    59746    238
50    10    156    671    113150    226


5    1    297    766    2548    509
5    20    171    781    25196    251
10    5    172    922    15418    308
10    20    156    813    46451    232
20    5    172    890    25684    256
20    10    141    813    45023    225
50    5    156    860    57672    230
50    10    156    812    130177    260



Clients    Users    Min    Max    Total    Avg
5    1    594    1016    3954    790
5    20    1172    3187    218414    2184
10    5    875    2219    77668    1553
10    20    1125    4485    587238    2936
20    5    1109    3391    216036    2160
20    20    938    7438    1732304    4330
50    10    578    8438    2461820    4923
100    8    1671    28593    9677890    12097

Hardware Configuration running JR:
- Intel Xeon 5130 (QuadCore) 2.0 GHz processor
- 3 GB RAM
- Windows XP Professional (32-Bit)
- Sun JDK 1.5.0_07
- JackRabbit 1.2.1
- Lucene 2.1.0
- JCR 1.0

Hardware Configuration running Oracle 10g
- Intel Xeon 5130 (QuadCore) 2.0 GHz processor
- 3 GB RAM
- Windows XP Professional (32-Bit)
- Sun JDK 1.5.0_07
- Oracle 10g as database for repository

Test Data:
Test data has 225 clients fullupgrade information in Oracle Database.
The Search Index for lucene is stored on the local machine.

A Customer has the following hierarchy:

/Product/Customer1/Configuration
/Product/Customer1/SalesData
/Product/Customer1/OtherData
/Product/Customer1/Configuration
/Product/Customer1/SalesData
/Product/Customer1/OtherData

Each of the above has some 30 to 50 nodes with properties.

Test Scenario:
Test is conducted for different clients (Customer1-Customer225) with
different combination of simultaneous users. Each user is a concurrent
thread. Each user is assumed to query 3 different subsets (nodes
within SalesData/Configuration/OtherData).
For example: In a setup when 10 customers are accessed by 5 concurrent
users each, 10x5x3 subset calls are made to JackRabbit.

Sequential / Random Customer Access:
In sequential Customer access, we request customers specific
information in order. If a test run needs to process 10 customers,
then it will query Customer1 through Customer225. In Random client
access from 225 clients, any 10 clients would be accessed.

Each sheet displays statistic for a single run.

Read/Write/Read test scenario tests concurrent Read/Write capabilities
of JackRabbit. Each user will have
3 queries. For the 1st and 3rd query, the program reads the data and
in the 2nd query it updates the node
with a test property..

The code:



              String clientId = "Customer" + new Random().nextInt(225);

              String[] xpaths = new String[]{
                      "Product/" + clientId +
"/Configuration/*/[EMAIL PROTECTED]'primary'and @name='connectivity']",
                      "Reno/" + clientId +
"/Sales/*/[EMAIL PROTECTED]'local'and @name='Bill']",
                      "Reno/" + clientId +
"/OtherData/*/[EMAIL PROTECTED]'admin'and @name='Carl']"};

              long before = System.currentTimeMillis();
              String nodeName = null;
              for (int i = 0; i < xpaths.length; i++) {
                  String xpath = xpaths[i];

                  if (i != 1) {
                      // Write a node
                      Query query =
_session.getWorkspace().getQueryManager().createQuery(xpath +
ORDER_BY_JCR_SCORE_DESCENDING, Query.XPATH);
                      QueryResult queryResult = query.execute();
                      NodeIterator iterator = queryResult.getNodes();
                      queryResult = query.execute();
                      iterator = queryResult.getNodes();
                      while (iterator.hasNext()) {
                          Node node = iterator.nextNode();
                      }
                  } else {
                      Query query =
_session.getWorkspace().getQueryManager().createQuery(xpath +
ORDER_BY_JCR_SCORE_DESCENDING, Query.XPATH);
                      QueryResult queryResult = query.execute();
                      NodeIterator iterator = queryResult.getNodes();
                      int itCount = 0;
                      while (iterator.hasNext()) {
                          Node node = iterator.nextNode();
                          nodeName = node.getName();
  //                    System.out.println(nodeName);
                      }
                  }
              }

Notes:
1. The above code runs within a thread.
2. Between creating threads, we have introduced a 1000 ms delay, and
we get the above results.
3. If we do not introduce the thread sleep, then the java process
quickly runs out of heap space for new threads.
4. Each time a thread starts, we logon.
5. Each time work completes in a thread, we logout.

-- Sriram

Single Workspace Single Repository (Oracle 10g) 225 Client Data DB Size 5330 MB Index Size 947 MB With Sequential Client Access 100 Millisecond sleep between each thread creation Clients Users Min Max Total Avg With Random Client Access 100 Millisecond sleep between each thread creation Clients Users Min Max Total Avg With Random Client Access 100 Millisecond sleep between each thread creation Read-Write-Read for one subset each in each user (thread) final String ORDER_BY_JCR_SCORE_DESCENDING = " order by @jcr:score descending";

smime.p7s
Description: S/MIME Cryptographic Signature