"The LDAP Guys."


Home        About        Pricing        Downloads        Key Relationships        Blog        Forums        Release Notices        Contact


Linux vs Solaris comparison of "CDS Silver" performance

10,000 Entry LDIF Scenario

The question has often been asked, "Which is the best operating system on which to run OpenLDAP?" This report is one of a series that look at OpenLDAP performance on different operating systems and hardware.

For this report we used systems that are categorized as high-performance, entry-level servers. The Sun V210 and Dell 1750 are two systems that come in similar configurations and are commonly used to provide LDAP directory services. We do need to point out that one system, the Dell 1750, boasts substantially higher processor clock speeds (3 GHz) than the Sun V210 (1 GHz). This factor very likely had a substantial impact on the performance difference we found between the two systems. However, it is also important to note that cost of the Dell 1750 was substantially lower than that of the Sun V210 system.

For quick study, the results are listed first. A detailed explanation as to the Directory Server configuration follows the results. A cost analysis based on the system prices is explored at the end of the report, as well as some further considerations.

Graphical Messaging Test Results
Graphical Addressing Test Results
Graphical Bulk Loading & Indexing Time Results

Summary of Test Results

  1. The Linux/x86 platform outperforms Solaris in all of the tests. The difference is particularly dramatic with the multiple client messaging test.
  2. Bulk Loading & Indexing of the database is substantially faster on the Linux/x86 platform.

Messaging Tests

Messaging Test with 1 client querying

Messaging Test with 10 clients querying



Addressing Tests

Addressing Test with 1 client querying

Addressing Test with 10 clients querying

Bulk Load & Indexing

Bulk Load Test with Indexing

Directory Server Configuration

CDS was configured as follows.

Setting CDS
Database Cachesize 384 MB
Entry Cachesize 10,000
IDL Cachesize 30,000
Access Logging Disabled

Cost Analysis

The following Cost Analysis is based soley on the cost of the systems. It does not include other costs such as: Support contract(s), System Administrator time, housing of hardware, etc. All amounts are U.S. Dollars.

The basic cost of the servers used is graphically visible here. Obviously, the Sun SunFire V210 Hardware is a bit more costly than the Dell 1750 system.

Using the above system prices, we can determine how many Operations/Second you recieve per dollar spent. The following graphs use the 10 client seeded average for each test type, and then divide that result by the cost of the system.

Using the above system prices, we can determine how many Operations/Second you recieve per 1,000 dollars spent. The following graphs use the 10 client seeded average for each test type, and then divide that result by the cost of the system divided by $1,000.

These graphs dramatically show the value gained by using the Linux OS on x86 hardware for Connexitor Directory Service.

Further considerations

Something to keep particularly in mind when evaluating the Linux vs. Solaris results is that for Solaris the database had to be directly mapped into shared memory via the shm_key configuration parameter. If shared memory is not used on Solaris, the time to load the database increases substantially and performance results drop significantly. With Linux, it has no effect either way. This essentially means that your Solaris system must have enough memory available to hold the DB environment and the running slapd process's memory footprint to function adequately.

 
 


Home        About        Pricing        Downloads        Key Relationships        Blog        Forums        Release Notices        Contact


     Copyright © 2001 - 2007, Symas Corporation.All rights reserved. The content on this page has been approved by Jordan H. Heyman, the Dog in "six guys and a dog."

     No rabbits were harmed in the production of this Web site.