Amazon Offering SSD-Backed Instances for Developers

Amazon is offering SSD-backed virtual servers to its cloud-computing clients, potentially speeding up the latter’s data-intensive applications. Solid-state drives, which lack the moving mechanical components of magnetic disks, have become increasingly popular as laptop and mobile-device storage.

“Modern Web and mobile applications are often highly I/O dependent,” read a July 18 posting on the Amazon Web Services Blog. “They need to store and retrieve lots of data in order to deliver a rich, personalized experience, and they need to do it as fast as possible in order to respond to clicks and gestures in real time.”

Amazon’s new “family” of EC2 instances, it added, “are designed to run low-latency, I/O-intensive applications, and are an exceptionally good host for NoSQL databases such as Cassandra and MongoDB.”

Amazon’s first offering in the family is the High I/O Quadruple Extra Large instance. Specs include 8 virtual cores, 60.5 GB of RAM, HVM and PV virtualization, 10 Gigabit Ethernet connectivity with support for cluster placement groups, and 2 TB of local SSD-backed storage (offered up as a pair of 1-TB volumes).

SSD storage is local to the instance. Amazon initially rolled out High I/O Quadruple Extra Large instances in its US East (Northern Virginia) and EU West (Ireland) regions, “at an On-Demand cost of $3.10 and $3.41, respectively.’

In a July 19 blog posting, Amazon CTO Werner Vogels wrote: “The SSDs will give you very high IO performance: for 4k random reads you can get 120,000 IOPS [input/output operations per second] when using PV and 90,000 when using HVM or Windows.” Write performance for SSDs is apparently more variable, depending on factors ranging from free space on disk to the type of file-system. “With PV virtualization we are seeing between 10,000 and 85,000 IOPS for 4k random writes and with HVM between 9,000 and 75,000.”

Databases in particular, he added, “can benefit tremendously from high performance I/O.” That’s because the I/O requirements for database engines, whether relational or non-relational, are notoriously demanding. Randomized access and burst I/O through aggregation can strain systems even further, particularly in NoSQL database management systems that back many scalable Web applications.

Those systems “require high replication factors to get to the aggregate random IO they require,” Vogels wrote. “Early users of [Amazon’s] High I/O instances have been able to reduce their replication factors significantly while achieving rock solid performance and substantially reducing their cost in the process.”

Meanwhile, the Netflix Tech Blog offers an extensive breakdown of how running Amazon’s High I/O instances apparently translated into faster speeds and lower costs.

SSDs: They’re not just for laptops anymore.


Image: Amazon