|By Unny Menon||
|September 22, 2013 03:00 PM EDT||
Increasing the I/O performance of next-generation data centers and cloud computing environments is essential for supporting enterprise e-commerce, database and email applications. To successfully deploy high-performance storage, it's essential for storage appliance software to be flash-aware and maximize solid-state drives (SSDs) for performance and hard disk drives (HDDs) for capacity, enabling an optimum price-performance storage solution.
With the explosive growth in and demand for information, next-generation data centers and cloud deployments require cost-effective storage solutions that provide end users with a high quality of service (QoS) in accessing and manipulating data. Usually, data centers - either corporate or cloud-based - run specific applications, such as database indexing and email exchanges. To provide a top QoS, these applications typically run on servers with high input/output operations per second (IOPS) and in turn provide access to data stored within the storage system.
As a general rule, the greater the number of users the higher the IOPS required from the storage system.
In this scenario, the I/O capacity will run out well before the storage capacity, so data center and cloud applications are required to add additional storage to aggregate performance. Storage systems have multiple applications working in the background to replicate data or manage its consistency. This practice results in low storage utilization and a complex data management scheme that increases the cost per user and cost per transaction of the storage deployment. Adding storage to scale for performance increases both capital and operating expenses, including power and cooling costs, and space requirements.
In general, SSDs can be used to alleviate performance bottlenecks within storage systems. Intelligently deploying SSDs to accelerate performance while also using HDDs for "cold storage" lowers systems' overall cost per gigabyte. This can be achieved by using file systems and software applications that are flash/SSD-aware. Modern file systems such as the Zetta File System (ZFS) are flash/SSD-aware and use SSDs to accelerate overall performance of the storage system.
ZFS is an enterprise-class file system that includes a built-in logical volume manager, along with reliability, availability and serviceability (RAS) features such as RAID-5. ZFS uses SSDs for two purposes:
- As a low-latency write log (ZIL-ZFS intent log) to improve overall storage system performance
- As a high-IOPS read cache (L2ARC) to improve "hot data" placement and primary data access latency
The ZFS Hybrid Storage Pool (HSP) is a robust, scalable file system with features not available in other file systems today. One of its revolutionary features is that it allows you to combine DRAM, SSDs and HDDs into an accelerated storage medium (see Figure 1).
Adaptive replacement cache (ARC) lives in DRAM. It's the first destination for all data written to a ZFS pool and the fastest (lowest-latency) source for data read from a ZFS pool.
Figure 1: An HSP using ARC
When data is requested from ZFS, it first looks to the ARC; if it is there, it can be retrieved extremely fast (typically in nanoseconds) and provided back to the application. The contents of the ARC are balanced between the most recently used and most frequently used data.
Level-two ARC (L2ARC): the L2ARC lives in SSDs and, in concept, is an extension of the ARC. Without an L2ARC, data that could not fit in the ARC would have to be retrieved from HDDs when requested.
SSDs that best match the write-log and read-cache requirements of modern file system deployments will result in direct savings of the performance and storage capacity. In other words, any improvement in the performance of the storage system's I/O capabilities will increase the number of users per system and will reduce the cost per user.
SSDs such as the sTec 3.5" ZeusRAM or 2.5" s840Z are ideally suited to fulfill the write-log requirement, while SAS SSDs like the sTec s800 Series meet the read-cache demands, enabling industry-standard storage to be converted into cost-effective, high-performance storage appliances.
Using a RAM SSD as write-log (Zil) and a SAS SSD as read-cache (L2ARC) can improve overall storage system performance up to 500 percent while supporting 15x number of users when compared to an HDD-only deployment, thereby reducing the cost per user and per transaction.