- Howard Chu
Carrier-Grade Stability and Performance
Early last year our client Acision ran a torture test on Symas OpenLDAP as part of acceptance testing for their telco offerings. These tests were performed on a directory consisting of one master replicating to two consumers, and included a number of crash/recovery tests as well as performance measurements. The short summary of the results: after two weeks of continuous testing, no problems.
Of course, things weren’t so rosy at the outset, and we had a few issues to investigate and correct at the beginning. But all’s well that ends well:
From: “XXX Petr” <XXX@acision.com> Subject: Stability tests on MDB after fix on masters Date: February 18, 2013 4:44:32 AM MST To: “Matt Hardin” <MHardin@symas.com> Hi Matt, The team has concluded the stability tests on MDB with no new issues – after two weeks of provisioning and a bit of torturing the both nodes were fully synchronized. Both issues of the following issues did not appear and are fixed from our perspective (again a big thank you to Symas!): Symas #1792 Excessive disk usage on both masters after restarts (sometimes) Symas #1790 syncrepl problem - number of records on consumer higher than expected Attached are the results of our stability tests on MDB + performance results. Please feel free to share the data with any external parties, or the community at large, as you deem fit. Best Regards, Petr
The attachments referenced in the email are available on https://symas.com/documents/20130218-Acision/ and consist of two Excel spreadsheets with raw data for the replication and torture tests, and one PDF summarizing the replication performance testing.
The key takeaway from these results - replication in OpenLDAP works reliably and efficiently, with extremely low CPU consumption. The software is crash-proof, and explicit attempts to kill the servers (in rude and messy ways) yielded no corruption or data loss. In fact Symas OpenLDAP has been in use by a number of telcos in recent years, and our software has never crashed in those deployments. When our software is properly configured it is completely rock-solid and the only reason for a service outage is actual hardware failures. No other software vendor in the world can truthfully make this claim.
To put this testing into context, a bit of historical background may be in order… LDAP is an integral part of the 3GPP Technical Specifications; these are the international standards behind all mobile telephony in use in the world today. At the heart of every telephone network is a set of LDAP directory servers maintaining data about every phone subscriber, handset, and network service. These directory servers must be ultra-reliable and ultra-efficient. They must routinely handle transactions for tens to hundreds of millions of users; telcos operating in Asia must scale to billions of users. If these servers fail then phone service across an entire country may fail. Fortunately OpenLDAP handles these requirements with ease.
While the rest of the software world struggles with reliability and scalability issues, Symas continues to quietly deliver solutions with unmatched reliability and efficiency.