Re: Cluster generating heavy traffic

[Mikael Ronström]

Hi,
From your description it is difficult to know what query that generates the
load. I guess you are reading the table every now and then. A few updates
per second is definitely not the problem here.

SCI will improve communication costs significantly yes.

Rgrds Mikael

Roberto Mendez wrote:
> I currently run a popular website that generates a
> lot of traffic 600+ mbit/s. I moved to a MySQL
> Cluster hoping that it can solve some issues for
> me such as performance, redundancy, scalability.
>
> I have a cluster that is made up of 3 machines, 1
> machine for the mgmt_node the other 2 machines are
> running ndbd and the mysql server.
>
> I have a simple table that contains about 90,000
> records that gets updated a few hundred times a
> minute.
>
> CREATE TABLE `security_keys` (
> `authkeys` text,
> `insert_time` int(11) unsigned default NULL,
> KEY `insert_time` (`insert_time`)
> )
>
> it is a very simple table that contains a string
> of charachters that get generated and deleted
> often. However once this table reaches about
> 90,000 records it generates traffic in excess of
> 400 mbit/s between the two servers that have the
> ndbd process running. The machines then get really
> slow (3.x+ load average) because of all the
> interrupts/context switches that are generated
> from all the network traffic being recieved on the
> NIC's. Leaves the system also waiting on I/O. If I
> remove all the records or never let it get above
> 1,000 records the traffic is next to nothing
> 2mbit/s however if I let it run normal I get all
> those issues.
>
> I am currently just testing the cluster system to
> see if it will scale to what I need. I will be
> adding some tables to the system that will contain
> 300,000+ records and be updated 100's of times a
> minute. Is this traffic normal between the two
> machines? If so will purchasing the SCI cards and
> SCI switches from dolphinics help out? Is anyone
> else seeing this much load on the systems and
> issues?