ZEA, A Data Management Approach for SMR

Digital data is projected to double every two years creating the need for cost effective and performant storage media [4]. Hard disk drives (HDDs) are a cost effective storage media that sit between speedy yet costly flashbased storage, and cheap but slower media such as tape drives. However, virtually all HDDs today use a technology called perpendicular magnetic recording, and the density achieved with this technology is reaching scalability limits due to physical properties of the technology [17]. While new technologies such as shingled magnetic recording (SMR) that further increase areal density are slated to enter the market [6], existing systems software is not prepared to fully utilize these devices because of the unique I/O constraints that they introduce.

SMR requires systems software to conform to the shingling constraint. The shingling constraint is an I/O ordering constraint imposed at the device level, and requires that writes be sequential and contiguous within a subset of the disk, called a zone. Thus, software that requires random block updates must use a scheme to serialize writes to the drive. This scheme can be handled internally in a drive or an alternative approach is to expose the zone abstraction and shingling constraint to the host operating system. Host level solutions are challenging because the shingling constraint is not compatible with software that assumes a random-write block device model, which has been in use for decades. The shingling constraint influences all layers of the I/O stack, and each layer must be made SMR compliant.

In order to manage the shingling write constraint of SMR HDDs, we have designed a zone-based extent allocator that maps ZEA logical blocks (ZBA) to LBAs of the HDD. Figure 1a depicts how ZEA is mapped onto a SMR HDD comprised of multiple types of zones, which are described in Table 1. ZEA writes logical extents, comprised of data and metadata, sequentially onto the SMR zone maintaining the shingling constraint.