Page 1 of 1
mysql ndbcluster slooooow on complex queries
Posted: Mon Sep 05, 2005 2:40 pm
by ody
I spent a good deal of time building an ndb cluster thinking being a memory based storage engine would be shiz fast.. well turns out anything more complicated than a single table select results in performance loss of 100x+ that of a simple innodb / myisam standalone server.
Has anyone setup an ndb cluster and noticed the poor performance of complex queries? or am just a simple minded town idiot thats missed something obvious (
d11wtq | 
).
cheers
ody
Posted: Mon Sep 05, 2005 7:14 pm
by Buddha443556
Connection speed?
Posted: Tue Sep 06, 2005 2:58 am
by ody
All nodes are on 100mbit full duplex connections into a cisco switch.
Posted: Tue Sep 06, 2005 3:13 am
by s.dot
My experience with queries and what will slow them down:
1. If they're inside of a for/while/foreach/list loop
2. Using joins, I've only used a join once, but it was 20 times slower than a regular select query
Posted: Tue Sep 06, 2005 10:58 am
by Buddha443556
Text indexes are not supported. Range statistics are not available. USE INDEX and FORCE INDEX will help that. You lose query caching too. No foreign key constraints.
I'm thinking NDB MySQL cluster isn't exactly meant to provide increase proformance but fail safe reliability.

Posted: Tue Sep 06, 2005 2:10 pm
by ody
Buddha443556 wrote:Text indexes are not supported. Range statistics are not available. USE INDEX and FORCE INDEX will help that. You lose query caching too. No foreign key constraints.
I'm thinking NDB MySQL cluster isn't exactly meant to provide increase proformance but fail safe reliability.

I was thinking the same thing, it seems to be more suited to situations where you need reliability and load handling.. mysql replication here I come!
Posted: Tue Sep 06, 2005 5:05 pm
by Buddha443556
ody wrote:I was thinking the same thing, it seems to be more suited to situations where you need reliability and load handling.. mysql replication here I come!
Hope you come back and tell us how that goes.
Posted: Mon Nov 14, 2005 2:46 pm
by ody
Yeap, replication works just as I needed.

Posted: Tue Nov 15, 2005 10:01 am
by BDKR
Buddha443556 wrote:
I'm thinking NDB MySQL cluster isn't exactly meant to provide increase proformance but fail safe reliability.
Exactly.
However, I think it's not that difficult to get to an exteme level of reliability without too much work with replication. I've use replication in two different clusters (one based on Turbo Cluster 6 and the other using Cisco Local Directors hardware load balancers) and the only thing that's tough to overcome is automated fail over when a master takes one in the gut. The first solution I had to manage this worked somewhat but I later realized that there was a much easier way to do it. Henceforth, my next project to
finish (I've allready started on it and gotten some way in, but I need to step back and take a 10k foot veiw then refactor).
Anyway, if you use only two boxes (master, slave), then at least make sure both have two mirrored drives at the very least. In that way, as far as I'm concerned, you have 4 seperate copies of your data set.
Good luck.