Idea ID: 2783716

Better Profiling and Tuning Tools for eDirectory

Status: Waiting for Votes

Waiting for Votes

See status update history

When troubleshooting eDirectory slowness (but not outright dysfunction), it is often very difficult to diagnose the cause. Even Novell Support often has difficulty pinpointing the exact cause of slow synchronization and directory request latency. Manual analysis of trace logs is often required.

Proposed Solution:
Existing tools such as iMonitor, the Novell Support Advisor, eDirMon, and ndstrace could be extended or new tools could be developed. Desired functionality could include:

1. Statistics on the type of access being requested to eDirectory (LDAP, NDAP, NCP, SYNC, etc.) including speed of transactions, relative resource usage per type of transactions, and overall volume of each access type. This seeks to answer questions like: Am I seeing too many LDAP requests on this server whereas NCP access is relatively low? Do I have too many partitions in my tree that are causing high latency versus consolidating partitions to minimize janitorial overhead?

2. System resource tracking: More meaningful information on how system resources (Memory, RAM, Disk, Network Bandwidth) are being consumed by eDirectory and whether eDirectory is constrained by any of them (and if so, by which). Also helps in determining cache size and whether file system caching should be preferred over native eDirectory caching (as with using Reiser and Ext3 on Linux).

3. Potential bottlenecks encountered by eDirectory and especially a metric of which bottlenecks are larger and smaller relative to each other. This helps to identify whether the administrator is attempting to optimize the wrong parts of eDirectory.

4. Comparative analysis of servers in a replica. Is one server causing the whole replica ring to slow? Is there load-balancing that can be done for some requests from one server to another?

5. Processing by process: Are my obituaries slow because directory writes are starving janitorial processes for time?

6. Improved obituary tracking: If I have 80 servers with backlinks, could the obituary reporting tools tell me which servers DO NOT seem to have been notified regarding a move, versus trying to list all the ones that have and determine which are missing? Between ndstrace and iMonitor, there is good obituary information available, but more could be added and aids to analysis and interpretation would be especially welcome.

7. Analysis of trace files: Tools to make it easier to determine trends from files generates with dstrace, rather than having to analyze data by eyeballing it.

8. Presentation of eDirectory performance data via graphs.

Value Proposition:
Much faster eDirectory analysis and troubleshooting. Performance tuning is currently the most difficult part of my eDirectory administration because my environment contains 80 partitions across nearly 100 servers with mixed platforms.