New arrays with a mix of SAS and SATA drives let you set up tiering in a single box. But proceed with caution: Numerous inoperability issues need to be resolved.
The emerging class of mixed SAS and SATA storage systems could be the next big disruptive technology. Mixing high- and low-cost SAS and SATA disk drives within the same system, at interconnect speeds comparable to Fibre Channel (FC), is a recipe for significant change and opens the door to data lifecycle management (DLM).
Most major storage vendors will release mixed SAS and SATA disk drive storage systems by early to mid-2008. The price of these systems is expected to drop below that of mixed FC/SATA arrays, and storage vendors will be jockeying for position. A spokesperson for Network Appliance (NetApp) Inc. says combining SAS and SATA drives in the same system is critical to the company's future product lines.
However, issues unique to SAS/SATA mixed-drive products are starting to emerge. For example, Overland Storage Inc. recommends users separate SAS and SATA disk drives in its new Ultamus RAID 1200 because the vibrations caused by the higher speed 15,000 rpm SAS disk drives cause slower spinning 7,200 rpm SATA disk drives with lower vibration tolerances to fail. Winchester Systems Inc. had to withdraw its FlashDisk HyperSAN SF-2300 Series SAS and SATA storage systems from the market partly due to the negative experiences that resulted from mixing SAS and SATA disk drives.
To address these issues, vendors are writing firmware code and testing specific disk-drive layouts in shelf drawers to minimize interoperability problems. Some vendors are introducing clustered and modular storage system designs that better exploit the inherent benefits mixed SAS and SATA storage systems can deliver. The degree to which users can capitalize on this mix of SAS and SATA drives varies in direct proportion to the management software vendors provide to migrate data between the different tiers of disk.
The number of disk drives, and the different combinations of how to configure them, is growing steadily. What follows are some key differences among Fibre Channel (FC), SAS and SATA disk drives, and why you may want to choose one over the other.
Early struggles and workarounds
Most vendors are candid about their early struggles and equally forthright about what users need to do to correct the situation. Winchester Systems found that within its first generation of products there were some basic issues that caused disk drives to logically drop out of an array group. In an attempt to fix this serious issue, Winchester Systems tried disk drives from different vendors but experienced the same problems. As a result, for some of its customers who had unresolved issues, Winchester Systems replaced their SAS and SATA storage arrays with FC systems.
Excel Meridian Data Inc. had better success by using only Seagate Technology disk drives (Cheetah SAS and Barracuda ES SATA) in its SecurStor Astra ES, which allows users to mix SAS and SATA disk drives any way they want. Excel Meridian found Seagate's firmware does a better job handling error conditions that occur when mixing its SAS and SATA disk drives than other disk drive manufacturers' firmware.
Overland Storage's dual-controller Ultamus RAID 1200 supports disk drives from different manufacturers, but users need to follow best practices to minimize performance and interoperability problems. For instance, Overland Storage recommends users assign SATA disk drives to one controller and SAS disk drives to the other; when storage is assigned, you can look at which controller is presenting the storage to determine what the performance characteristics are on the volumes being presented.
Another best practice recommended by Overland is to install all of the SAS disk drives in the same vertical columns and all of the SATA drives in different vertical columns. Overland found that when users put SAS disk drives in rows across the array instead of columns and then create volume groups across these rows of disks, the combined rotational speed and shock of this group of SAS drives generates excessive vibration, which can affect the reliability of SATA drives in adjacent rows.
EqualLogic Inc. also supports SAS and SATA disk drives from multiple vendors within its PS Series storage systems, but the different class drives can't reside in the same system. However, EqualLogic's virtualization software lets you create a single storage pool that allows up to eight storage arrays (called members) to act as a single logical storage system. EqualLogic also includes performance-tuning software that load-balances hot spots across the SAS and SATA storage array members that are part of the larger storage system. This allows administrators to grant servers whatever tier of storage application owners need and then optimize the data's placement over time.
Most vendors expect to resolve most of the kinks related to mixing SAS and SATA in a single storage array by the end of this year. They anticipate that most of the current interoperability issues will resolve themselves as SAS protocols and firmware on disk drives and storage controllers matures.
One of the key benefits of mixed SAS and SATA architectures is that the same disk-drive socket can accommodate SAS or SATA disk drives. This differs from the current generation of mixed FC and SATA storage arrays that don't permit the insertion of SATA disk drives into FC sockets or vice versa. Using both types of drives in the same socket allows users to grow performance or capacity using existing free sockets in the storage arrays while simplifying the planning process around obtaining more sockets for disk drives.
The SAS protocol also possesses characteristics that, while not as strong as FC, are satisfactory for the demands most enterprises place upon their storage arrays. For instance, switched FC supports 4Gb/sec speeds and can address more than 16 million devices, while SAS offers a switched 3Gb/sec back end and can address 4,032 devices per port. Even though SAS speeds and the number of addressable units are lower than those for FC, they're sufficiently large to allow mixed SAS and SATA storage arrays to grow to meet most enterprise requirements for speed and capacity.
NEC Corp. of America's D-Series storage line is one of the first to capitalize on the benefits SAS offers. The D-Series is available in six models from the entry-level 72-drive 54TB D1-10 to the enterprise 1,536-drive 1.15 petabyte (PB) D8-1040. Being fully modular, the D-Series lets users upgrade from the D1-10 model to the D8-1040 model without a rip-and-replace strategy. SAS makes these types of modular configurations possible and, in so doing, helps to eliminate the need to deploy different storage models for midrange and enterprise requirements and recreate storage islands.
Products like Dot Hill Systems Corp.'s 2730T, Excel Meridian's SecurStor Astra ES and Overland Storage's Ultamus RAID 1200 similarly capitalize on the SAS protocol's ability to address larger numbers of disk drives, as these models include expansion packs that let users grow each of their base units from 12 disk drives to 60 disk drives. However, these are only JBODs and they don't allow users to introduce more controllers to handle additional performance requirements, although with Excel Meridian you can upgrade to a more powerful controller. This is where NEC maintains a distinct advantage for now, as users can scale its D-Series up to 16 controllers on NEC's D8-1040 to handle the increase in performance required by the introduction of more disk drives. Yet what NEC's D-Series and most other storage systems in this class don't yet accomplish is automatically placing data on the most appropriate tier of disk.
The real challenge and opportunity for mixed SAS and SATA storage systems is the ability to move data between different tiers of disk within a single storage system; higher performing data is placed on SAS drives, while data with more modest performance requirements is put on SATA drives. But to place the data on the right volumes requires administrators to know which ones are SAS volumes and which ones are SATA volumes.
To accomplish this on most SAS and SATA storage systems will require some third-party tool such as TeraCloud Corp.'s TeraCloud Storage Framework (TSF) Lite, which monitors the utilization and growth of data in your environment. As data approaches specified age, capacity or performance thresholds, the tool alerts administrators to these issues.
When these threshold issues occur, administrators will likely still need a way to migrate data from one volume to another within the storage system. Some of the SAS and SATA storage vendors, such as EqualLogic, Hewlett-Packard (HP) Co. and NEC, include replication software to replicate data from one volume to another.
If you need to move individual files, products like CommVault's QiNetix DataMigrator or Silicon Graphics Inc.'s (SGI) InfiniteStorage Data Migration Facility (DMF) allow administrators to set policies that monitor file use and growth, and automatically migrate files to the appropriate tier of back-end disk based on access frequency or file type. This approach requires administrators to install host agents, set up policies and administer the products on an ongoing basis, although SGI plans to release an integrated solution that will work with its DMF software and InfiniteStorage 220 storage array in late 2007.
EqualLogic allows users to automate the movement of data from one volume to another on their storage systems, but it only works under the following circumstances:
- Each PS Series storage array supports only one type of disk drive--SAS or SATA--so to introduce automated tiering of data, you must deploy at least two EqualLogic storage arrays with different types of disk drives.
- Automated data placement is based on the I/O characteristics of each specific volume. Administrators then set policies or accept the default policy levels so the EqualLogic software automatically moves the volume from one tier of storage to another. This setting is not granular to the file or block level.
- Both volumes must be in the same storage pool, although these storage pools can cross multiple storage arrays. You can only create up to four storage pools, so if volumes reside in pools outside of those volumes the scope of functionality may be limited.