Amazon takes the lead in providing high performance I/O in the Cloud

Thursday September 27, 2012 , 4 min Read

One of the concerns of moving workloads to the Cloud is the degradation in the I/O performance. Since multiple instances rely on shared storage, the I/O performance does take a hit. This becomes very obvious when the application heavily relies on disk I/O for the database transactions. Many AWS customers consistently raised the issue of poor I/O for Elastic Block Store (EBS) and Amazon Relational Database Service (RDS). One of the key themes for AWS in 2012 is enhanced I/O performance. In the last few months, many services of AWS have been upgraded to take advantage of this feature. The recent service to get this feature was Amazon RDS.

At the beginning of this year, Amazon went live with their new managed NoSQL service called DynamoDB. This service saves the customers from the hassles of setting up, scaling and managing a NoSQL database cluster based on MongoDB or Cassandra. DynamoDB was the first to introduce the concept of provisioned throughput where the customers can request the desired read and write throughput while creating the database. Amazon achieved this by backing the storage with SSD. This set the precedence for other I/O intensive AWS services to follow. The price is based on the factors of provisioned throughput, data storage and data transfer.

In July, Amazon announced High I/O instances for EC2 that came with very high, low latency, disk I/O performance using SSD-based local instance storage. Amazon replaced the default ephemeral storage with 1TB SSD for a type of EC2 instance called h1.4xlarge. This type of instance is great to run a NoSQL or a relational database that gets better with a faster disk I/O. These instances also come with high network I/O based on Gigabit Ethernet making it ideal to run a Apache Hadoop cluster on the Cloud. This service was the second one to be based on SSD from AWS.

During the same month, AWS offered Elastic Block Store (EBS) with support for provisioned IO Operations per Second (IOPS). This EBS volume type was designed to deliver predictable, high performance for I/O intensive workloads, such as database applications, that rely on consistent and fast response times. Customers can attach multiple volumes of EBS and stripe them further to achieve higher IOPS. Along with this Amazon introduced the ability to launch special EC2 instance types called EBS-Optimized instances to deliver dedicated throughput between Amazon EC2 and Amazon EBS, with options between 500 Megabits per second and 1,000 Megabits per second depending on the instance type used. Synchronizing data between the SSD based ephemeral storage and the EBS with provisioned throughput gives the required performance, reliability and durability of data.

This week Amazon has announced the availability of Amazon RDS with provisioned IOPS that enables customers to provision up to 10,000 IOPS with 1TB of storage. This addresses all the performance concerns raised by RDS customers by making it an ideal database service on the Cloud to run high performance OLTP operations. At this point, customers can only go up to 10,000 IO operations per second. Amazon says that will provide an automated way of migrating existing RDS databases to the provisioned IOPS storage in the future.

Amazon has been quick to respond to the customer needs by bringing high performance I/O to the Cloud. This certainly makes AWS the preferred Cloud to run I/O intensive tasks demanding high throughput of disk I/O, network I/O and OLTP operations. Based on the pattern, I am expecting that the next service to get the provisioned throughput would be Amazon Elastic MapReduce which can get a huge boost in processing large datasets due to the enhanced network and disk I/O.

- Janakiram MSV, Chief Editor, CloudStory.in

Advertise with us