List:Cluster« Previous MessageNext Message »
From:Ph.D. Joseph E. Sacco Date:July 30 2004 4:01pm
Subject:Building a cluster with a slow and a fast machine: Redux...
View as plain text  
Systems:
* PowerMac with dual   G4 533MHz CPU's, 1GB RAM, SCSI drives
* PowerMac with single G3 240MHz CPU, 264MB RAM, SCSI drives
* Yellow Dog Linux-3.0.1: 2.4.25 kernel
* mysql-4.1 from BK tree 29July04

=============================================================
I have managed to get a two replica, four db node cluster running on two
machines with drastically different resource/performance characteristics
thanks to recent comments to the list from Mikael Rostrum.

The default DB configuration parameters used: 
[DB DEFAULT]
Replicas: 2
DataMemory: 40M
IndexMemory: 12M
HeartbeatIntervalDbDb: 5000               <<=========
TimeBetweenWatchDogCheck: 30000           <<=========

Increasing the TimeBetweenWatchdocCheck eliminated the type of db node
crash on the slower machine that showed up in the log files as:

        Ndb kernel is stuck in: Job Handling

Increasing the HeartbeatIntervalDbDb eliminated db node crashes arising
from too many "missed heartbeats". 

So... It works!

I am the first to admit that running a two replica, four node cluster on
these two systems is like trying to get a dog to dance on its hind legs.
To what end??? It is just a learning experiment.

For what it is worth... During the startup, the average load on the
small machine climbs above 5 as the amount of memory used by swapd
climbs to 300MB and the disk block I/O goes into coronary arrest.  Once
the cluster is stable, the average load drops to 0.6 and the block I/O
settles down.

-Joseph

-- 
Joseph E. Sacco, Ph.D. <jsacco[at]earthlink[dot]net>

Thread
Building a cluster with a slow and a fast machine: Redux...Ph.D. Joseph E. Sacco30 Jul