CODEWITHSUNDEEP
databasepostgresqlamazon-web-servicesioamazon-rds
hndr
hndr

Reputation: 767

Unusually low read IOPS on RDS

This is essentially the same as the unanswered question posted on the AWS forum here.

I'm running PostgreSQL on Amazon RDS, m3.large instance with 1000 provisioned IOPS, but I'm finding that the read IOPS is unusually low, much lower than the write IOPS although the application is a read-intensive application.

quick IOPS measurement

Read queries are also unusually low, with a simple query involving 2 tables taking up to 2 seconds, sometimes longer, even when limited to only 50 rows.

Where could the problem be? What can I do to improve the query performance?

Upvotes: 0

Views: 2399

Answers (1)

mootmoot
mootmoot

Reputation: 13166

This question should move to serverfault.com. Anyway, it seems many people jump into provisioned iops before digging deeper on RDS IO price/performance : Why buying Provisioned IOPS on RDS may be a mistake

Here is some common sense. (us-west-2)

  • General purpose SSD(gp2) $0.115 /GB , 3 iops
  • Provisioned iops $0.125/GB (3 iops) + $0.10 per iops

So what should you do to allocate 1000 iops and need to store 100GB of data?

  1. gp2 : 1000 /3 = 334GB , Cost = $38.41
  2. Provisioned IO : 100GB x 3 iops + 700 provisioned IO. Cost =$12.5+$70 = 82.50

Now the better part of gp2, If you didn't exhaust all the guarantee I/O, you can even burst over the 1000iops compare to provisioned io.

Next is the query performance. IMHO, this is a bad question : where is your query code proof? Did you benchmark your query results instead of watching the RDS iops chart? Did you use cloudwatch to records the iops trends?

Upvotes: 1

Related Questions