"The LDAP Guys."

Symas Benchmarking Home

LDAP has been the defacto standard for how users and applications access information stored in a directory server for over 5 years. On these pages the performance of Connexitor Directory Services (CDS) provided by Symas Corporation is compared across platforms and against other Directory products. A previous study can be found here, comparing various Directory Service technologies, but it dates back to May 15, 2000. The first set of comparisons for the different benchmarks is always set to mimic the 2000 benchmark for comparison sake.

A detailed explanation as to the benchmarking tools, hardware, Operating Systems (including tuning), and test descriptions.

CDS3 Authentication Rate Benchmarks

1. CDS 3 Authentication Rate Benchmarks (May - June, 2006) are comparable to recent benchmarks of Java Enterprise Directory Server (JSDS) by Sun.

CDS Hardware Comparison Benchmarking Reports

1. "CDS Silver" performance comparison of Linux and Solaris
   Updated February 5, 2005

Directory Server Software Benchmark Comparisons

1. "Netscape DS 6.21" vs "CDS Silver"
   Updated February 5, 2005

"Performance of Symas Connexitor Directory Services on HP BladeSystem servers"

1. This Hewlett Packard report (PDF) discusses a series of benchmarks using the slamd benchmarking tool to evaluate CDS performance on various HP BladeSystem configurations with varying directory sizes. August, 2006

Special Considerations

The benchmark numbers for a particular platform will change over time as the tests are performed on newer versions of the underlying software. Every effort will be made to bring older benchmarks up to date as time allows.

Tools used for this comparison

For these reports, DirectoryMark was used as the benchmarking tool when performing searches, as it has been widely used to get performance results on LDAP servers for many years. For the time-based metrics used for the bulk loading results, the unix utility "time" was used.

Hardware used in these comparisons

1. SunFire v210 from Sun Microsystems. Installed with two UltraSPARC-IIIi CPU's running at 1 GHz, 2 GB DDR RAM, and two Seagate 33.9GB SCSI Hard drives.
2. Dell 1750 from Dell Computers. Installed with two Intel Xeon CPU's running at 3.0 GHz with Hyper Threading, 2 GB DDR RAM, and two 36GB 10K RPM Ultra 320 SCSI Hard drives with RAID 1 mirroring.

Operating Systems

1. The SunFire v210 runs Solaris 8.
2. The Dell 1750 runs Debian Woody

Solaris was tuned following the suggestions from Sun's idsktune utility & other Solaris tuning guides.

NDD Tuning

• /usr/sbin/ndd -set /dev/tcp tcp_time_wait_interval 30000
• /usr/sbin/ndd -set /dev/tcp tcp_conn_req_max_q 1024
• /usr/sbin/ndd -set /dev/tcp tcp_keepalive_interval 600000
• /usr/sbin/ndd -set /dev/tcp tcp_ip_abort_cinterval 10000
• /usr/sbin/ndd -set /dev/tcp tcp_ip_abort_interval 60000
• /usr/sbin/ndd -set /dev/tcp tcp_smallest_anon_port 8192

/etc/system Tuning for Solaris

• set rpcmod:svc_default_stksize=0x6000
• set lwp_default_stksize=0x6000
• forceload: sys/shmsys
• set shmsys:shminfo_shmmax=1610612736
• set priority_paging=1
• set maxpgio=65536
• set slowscan=500
• set fastscan=64306
• set handspreadpages=30960
• set autoup = 60
• set tcp:tcp_conn_hash_size = 16384
• set rlim_fd_max = 16384

Testing Vocabulary

There are two types of search tests executed, and one data load with indexing (bootstrap) test. The two types of search tests are "addressing" and "messaging".

1. The addressing test involves doing queries against the directory server using a mix of exact & substring searches on different sets of attributes, where some of the queries are for entries that do not exist. This tends to emulate "phonebook" or "addressbook" type queries that come in from email clients and applications.
2. The messaging test involves doing exact match queries against existing entries using a single attribute. This tends to emulate "email delivery" type queries that come in when the email infrastructure uses LDAP lookups for email delivery.

Two levels of stress were applied with the addressing and messaging tests. In the first level, only a single client would query the server, gauging the server.s ability to best respond to that client. In the second level, 10 clients would query the server simultaneously, gauging the server’s ability to handle multiple queries.

A unseeded test is one where the server has just been started, and has no entries stored in its cache when the test is initiated.
A seeded test is one where the server has already had a set of queries performed against it, filling in the cache.

The Bulk Load & Indexing test looks at how long it takes to load the entry database into the Directory server and index it using the quickest method possible.