Scalability updates to the Amazon Web Services Elastic Block Store service became generally available this week, and while the changes are welcome, some users say the new volumes could be simpler.
The Amazon Web Services (AWS) Elastic Block Store (EBS) can now scale up to 16 terabytes (TB) per volume, up from 1 TB before; with provisioned IOPS, SSD-backed EBS volumes can now achieve up to 20,000 IOPS per volume as well. General Purpose (SSD) volumes can now scale to 16 TB as well, and deliver up to 10,000 IOPS with a maximum throughput of 160 Megabytes per second (MBps).
This is good news for users who have been dealing with striping data across multiple volumes for either performance or capacity reasons.
"The faster IOPS volumes will be immediately beneficial [because] they mean that we no longer have to stripe server volumes over multiple EBS volumes to get required throughput levels," said James Fogerson, solutions architect for Robert Half International Inc., a staffing firm in Menlo Park, Calif. "We've had some recent incidents where we needed more performance than was available in a single EBS volume."
Jared Reimerco-founder, Cascadeo Corp
With 1 TB volumes, users had to aggregate multiple volumes for larger volume sizes using software RAID, a process that could be complex and error-prone, according to Jared Reimer, co-founder of Cascadeo Corp., an IT consulting firm located in Mercer Island, Wash.
Software RAID can be complex to set up correctly, and setting it up incorrectly yields several problems. One common mistake, according to Reimer, is telling the logical volume manager to write sequentially across volumes, rather than striping across them uniformly.
"It is easier for a non-storage-expert to work with a single, large, performant volume than it is to ask them to correctly design, implement, and troubleshoot a software RAID volume," Reimer said.
Software RAID also adds another layer of indirection and could mean a performance penalty on writes.
"This depends on a number of factors but generally speaking, we'd prefer to avoid this if we can do so," Reimer said.
Hot spots in data storage are also problematic. This new approach appears to mitigate some of that impact by allowing a large number of aggregate IOPS across a larger AWS EBS volume, instead of hoping the software RAID strategy effectively stripes the data across a number of smaller, less-performant EBS volumes, Reimer said.
"Essentially it turns something that was previously a painful corner case into something routine and trivial to work with," Reimer added. "It's not going to help everyone, but the people that it helps will be very grateful to have it."
What would help further would be a simpler upgrade process to move to newer, more highly performant AWS EBS volumes, according to Kevin Felichko, CTO of PropertyRoom.com, an online auction company based in Frederick, Md.
Currently, users have to snapshot, create a new volume with the new IOPS setting, attach it to the instance, and detach the old volume, Felichko said. This prevents Felichko's team from doing some needed provisioned IOPS upgrades, because it would either require some downtime or additional cost and time to fire up replacement EC2 instances to replicate data.