Amazon EBS outtage

Recent outages of Amazon cloud at the end of August mark the third significant outage in two years caused by Elastic Block Store (EBS) failures.  According to Memset, who offer persistent local storage by default, these high profile outages suggests that network-attached block storage is not always a reliable place to store data. 

Juan Martinez, DevOps leader at Memset, said, "When you launch a cloud server with Memset, the virtual machine and storage are persistent, with data being stored on regular disks or SSD disks if you require high speed input/output. This storage is managed with traditional hardware durability strategies (RAID) and thanks to the abstraction layer provided by our cloud platform, it is completely transparent."
 
"Additionally we offer the possibility to take snapshots of this storage and store them inside of an object storage system based on clustering technology that guarantees three copies of each object in different failure zones. This is particularly useful for disaster recovery, storing multiple copies of your server or adding additional servers to an existing cluster or deployment," concluded Martinez.

Memset's Memstore cloud storage service, based on OpenStack SWIFT, has also proven extremely reliable with zero data loss to date. If there is a disk failure on a Miniserver VM host, Memset's operations team can swiftly change the affected drive in the hot-swap chassis meaning that data losses for their VMs are also vanishingly rare. There has been no loss of Miniserver VM data as a result of infrastructure or platform software failure in at least the last five years.

“Many of our customers prefer our Miniserver VM solution because of the persistent nature of our VM's storage. If you want low-latency, persistent storage for a VM little beats just using the VM host's local RAID1. Generally we don't often need to move VMs between hosts so a centralised storage system is largely unnecessary; we just copy the image across when we do.” says Kate Craig-Wood, MD of Memset.

Craig-Wood continued, “These outages suggest that some cloud approaches are not yet enterprise-ready. While Amazon's system is certainly very clever, our straight-forward approach seems increasingly a good move. This means a failure is unlikely to take out more than one of a customer's VMs at once, and also makes our Miniservers operationally identical to a dedicated server which our customers like. It is a classic application of the KISS engineering principal.”

-ends-