cache memory in I/O controller affecting performance in Vertica Server

I am 100 % sure that performance will be impacted.

Heathy nodes will be waiting for slower node to finish it's portion of processing.

Since Vertica does not have database buffer cache it needs to read data from hard drive for every query.

Degraded IO controller will have impact for sure.

You can approach impact assesment 2 ways.

1. PROFILE query and look at EXECUTION_ENGINE_PROFILES table. Look at "execution time (us)" counter and group by node. You should see it node with faulty controller slower.

2. Sometimes if you have degraded hardware it is better to have node DOWN then running. I know it may sound strange, but distributed systems have different approaches. With high node count having 1-2 nodes DOWN could be treated as normal operations. In our 48 node cluster we run fine for weeks with at least 1 node down. If you have small node count it may be not acceptable. You need to assess what portion of computing resources your cluster will loose if node will go down before doing this.

Hi Kruity

Thanks for the response. As per your suggestion, I trigger following queries but i could not understand You should see it node with faulty controller slower. How can i verify faulty controller slower ? where as cache failed has occured on node 4

1. Below is the query i trigger

select node_name,user_id,user_name,session_id,counter_name,counter_value from EXECUTION_ENGINE_PROFILES where node_name='v_testpulse_node4' and counter_name='execution time (us)' group by node_name,user_id,user_name,session_id,counter_name,counter_value

PLEASE ADVISE if query is wrong

2. I have attached the output on excel file. Please find the attached excel file for an output.

QUESTION :

1. How can i see the timestamp for faulty controller slower ? Since it has occured on july 15th onward till july 30th, 2016 and today is August 15th, 2016