List:Cluster« Previous MessageNext Message »
From:Mikael Ronström Date:August 30 2004 5:52am
Subject:Re: API-Nodes & memory usage on large systems
View as plain text  
Hi Michael,

2004-08-30 kl. 00.00 skrev Michael Spindler:

> Hi list,
> I tested MySQL-Cluster for quite a while now and there are some 
> questions left, hopefully someone clear those things up for me:
> - Am I right, that there is no automatic failover for API-Nodes at the 
> moment? Do I have to care about those API-Nodes myself, if I want high 
> avaiable systems and in best case some sort of load balanced queries 
> to my cluster (using LVS or similar systems)?

 From MySQL clients there is automatic failover using Connector/J, the 
work on MySQL C API's is still in the TODO section. The MySQL Servers 
has automatic
failover between the DB nodes and also load balancing.

> - Is there any special memory/disk/whatever configuration needed at 
> the API-Nodes, to get the most speed out of the cluster or do they act 
> just like a kind of relay?

If there are any parameters needed they are part of the cluster 
configuration at the moment (none sofar, will be a few in 4.1.5).

> - From MySQL-Manual:
> "...The default DataMemory size is 80000 kBytes. The minimum size is 1 
> MByte and the maximum size is slightly more than 3 Gbytes..."
> "...The default IndexMemory size is 24000 kBytes. The minimum size is 
> 1 MByte and the maximum size is around 1.5 GByte. ..."

The manual documents 4.1.3 where the max size was limited. This was 
fixed in the bk tree before the summer and will be part of the 4.1.4 

> If I would use a system with 16 GB of main memory and two processors, 
> I would think, that I should configure the maximum mentioned in the 
> manual. Using two DB-nodes on this server (to load both cpus), I would 
> get a maximum memory usage of 2x3 GB (data) + 2x1.5 GB (index) = 9 GB
> So, I would have to configure more than two DB-nodes for fully using 
> the main memory!? Wouldn´t that mean, that the hole system slows down, 
> when there are three or more processes sharing the cpus? I

With current bk tree you can handle the entire memory with one DB node 
per machine.

Rgrds Mikael

> Michael
> -- 
> MySQL Cluster Mailing List
> For list archives:
> To unsubscribe:    
Mikael Ronström, Senior Software Architect


API-Nodes & memory usage on large systemsMichael Spindler30 Aug
  • Re: API-Nodes & memory usage on large systemsMikael Ronström30 Aug