Showing posts with label TADDM performance VMware. Show all posts
Showing posts with label TADDM performance VMware. Show all posts

Sunday, March 15, 2009

Comparing TADDM performance on different platforms

Introduction


This document compares the running time of running discovery in TADDM 7.1.2 on a virtual environment (VMware ESX) and a non-virtual environment. This is not an comprehensive test, and should not take as TADDM performance result, but just as a comparison study of the TADDM Server in different platforms.

Platforms

Virtual environment


Linux RedHat Enterprise Server Release 5
2 vCPU
3600 MB
VMware ESX, with just 1 VM running

Non-virtual Windows environment


Windows 2003 SP2
2 CPUs
3200 MB
TADDM 7.1.2 and DB2 collocated in one server

Non-virtual Linux environment


Linux RedHat Enterprise Server Release 5
2 CPUs
3200 MB
TADDM 7.1.2 and DB2 collocated in one server

Two server, non-virtual environment


TADDM Server in the Windows environment above
DB2 Server in a separate machine, same specification as the server above


Target environment


37 Computer Systems, consisting of Windows, AIX, HP-UX, Solaris and Linux machines.
Number of components found: 315 (3 Server Equivalents)


Running time


Running time
Virtual environment
Windows environment
Linux environment
Two-server environment
Level 3 Discovery
30 min 47 sec
8 min 49 sec
28 min 37 sec
6 min 4 sec
Level 3 Discovery without WebSphereCellSensor
10 min 36 sec
7 min 43 sec
6 min 16 sec
5 min 8 sec
Level 2 Discovery
10 min 15 sec
5 min 6 sec
7 min 34 sec
4 min 4 sec
100 MB IDML bulk load (250,000 CIs)
13 hours 9 min
5 hours 52 min
13 hours 55 min
22 hours 19 min
Level 3 Rediscovery after bulk load
30 min 4 sec
6 min 47 sec
16 min 25 sec
6 min 24 sec


Conclusions


  • Even in a small scale, the discovery process in a Virtual environment is almost 4 times slower than in a non-virtual environment
  • Although the run time for other activities don't show significant difference, the discovery process in a Virtual environment is just prohibitive.
  • Except for bulk loading, a two-server environment had the best performance in all tests.