Until the costs drop even further, performance-boosting solid-state drives (SSDs) won't make economic sense for every type of application, so it's important to follow best practices to ensure they are working for your storage shop. Here are 10 SSD best practices to follow:
1. Identify I/O-intensive applications that will benefit from faster data storage.
Types of applications that may be well suited to SSD technology include databases, data mining, data warehousing, analytics, trading, high-performance computing, server virtualization, Web serving and email systems.
A check of how many enterprise-grade 15,000 rpm and 10,000 rpm hard disk drives are in use and how much money has been spent on DRAM for performance will help to determine if SSDs will be worth the investment.
Study application workloads and assess hot vs. warm vs. cold data sets. Active data can be directed to the flash solid-state drives, and the less frequently accessed data can go on Fibre Channel (FC) drives or SATA drives.
"If you have a good understanding of that, then you can understand how much solid-state storage you're likely to need to be able to optimize the performance of your system," said Jim Handy, an analyst who focuses on memory chips and SSDs at Objective Analysis.
2. Ensure that applications, especially those that are custom-written, can handle the faster solid-state drives.
"For most applications, this should not really be a problem, but depending on coding and timings, you can have the chance to have things done too quickly which can throw off timing a bit, as well as the processors actually jumping due to it not having to wait for the disk anymore," said Jon-Carlos Mayes, IT director at CCP hf, a Reykjavik, Iceland-based online game producer. CCP uses both DRAM and NAND flash SSD technology from Texas Memory Systems Inc.
3. When evaluating solid-state drive technology, concentrate on cost per IOPS, not cost per gigabyte.
"Focus on what would be the lowest overall system cost to get the throughput that you require," Handy said. "If you focus on cost per gigabyte, then a solid-state drive will always look bad because it ranges from 20 times [more than] the cost per gigabyte of a hard disk drive."
4. Make sure the performance and reliability of a vendor's SSDs can be measured in terms of random reads and writes across small blocks and pages.
"Vendors will quote you whatever they can do the best in the lab, and that may not be what you're actually running," said Joseph Unsworth, a research director at Gartner Inc. Once you determine which supplier can deliver the results you need, he added, have the vendor do a proof of concept and make sure your service-level agreement is tailored specifically to your application needs.
5. Determine which type of solid-state drive will be the best fit.
The chief SSD choice confronting users today is NAND flash or DRAM. DRAM SSDs are significantly faster and perform reads and writes equally well, but they're also considerably more expensive and consume more power. NAND flash SSDs -- whether single-level cell (SLC) or multilevel cell (MLC) – perform better on reads than writes, and they wear out over time.
"Are you just accessing data? Or are you doing a lot of programming of data? That will determine whether or not you can go with a cheaper but less robust multilevel cell-based SSD, or you need to go with single-level cell," Unsworth said. "If you're doing a mix of both [reads and writes], then you're going to want to make sure that you're using the single-level cell technology over the multilevel cell."
The more expensive SLC solid-state drives are better suited to the enterprise because the wear life is longer for continuous writing than it is with MLC-based drives.
NAND Flash SSDs can be especially helpful for a read-intensive database table, for example, whereas DRAM SSDs -- whether in an appliance, in cache or in DRAM combined with flash -- would be a better option for transaction logs or journal files, where you're recording a copy of what's changing, said Greg Schulz, founder and analyst at StorageIO Group.
"I tell folks if they're going to need to have super transaction rate capability because they're running some data mining application, then one of the high-performance boxes like Violin or Texas Memory Systems has could be a pretty economical way to go," said Gene Ruth, a senior storage analyst at Burton Group. "The alternative would be to build out a huge hard disk drive-based system with all the power and space and maintenance and failure rates that go along with having lots of hard disks."
6. Consider NAND flash solid-state drives for caching purposes as a way to augment application performance.
Write caching is typically done at the storage device with DRAM cache that persists the writes to back-end storage. When that approach isn't fast enough, DRAM solid-state drives have been used to accelerate write-intensive applications. Now that lower cost flash is available, SLC SSDs will increasingly become the preferred option for write caching, especially when budgets are tight.
The effectiveness of any read caching layer depends on the size of the data set being accessed, the frequency with which the data is read and the performance of the cache. If the data set is small and being read on a frequent basis, server RAM usually suffices. But if the data set is large and the reads are random, flash SSD can work well. Although SLC and the less expensive MLC are both options, the more durable SLC is generally favored.
7. Consider solid-state drive over short-stroking.
Formatting a hard disk drive so that data is written only to the outer sector of the disk's platter can increase performance in high I/O environments, since it reduces the time the drive actuator needs to locate the data. But that practice, known as short-stroking, leaves a substantial percentage of the disk drive's capacity unused.
"You're deliberately not using what you bought. Because we're so used to it, people think that's how it has to be," said Mark Peters, an analyst at Enterprise Strategy Group. Even though SSD technology is nascent, he said, it already makes financial sense in scenarios where users now short-stroke. "Everyone says, 'I don't want to buy solid state because it's 10 times the cost [of hard disk drives],'" Peters said. "But you could take that 5 GB from 20 [hard disk] drives and put it on one solid-state [drive]."
8. Determine how much power your data center is consuming.
According to Peters, "You're either in parts of the country or a data center where you have oodles of power, or you're in parts of the country or a data center where actually you don't have much power left and you're going to hit the wall at some point. If you don't know how much [power] you're using, how can you know when that's going to happen?"
Because SSD technology is more energy efficient than hard disk drives, it can help to extend the life of a data center with power constraints. But, Peters said, many IT organizations have no clue what their electric usage is.
9. Experiment with solid-state drive technology in the lab.
"They definitely want to bring in, if not the individual disk devices, possibly a subsystem that's based on SSDs," Burton Group's Ruth said, possibly to target a particular application for test purposes. Even using SSDs in a laptop can help to illustrate the potential advantages, he added. "People are familiar with hard disks. They get that. They need to develop that comfort level with solid-state disks as well."
10. Make sure long-term planning takes into account a potential solid-state storage tier.
Solid-state drives will be integrated into storage systems as standard fare going forward, so IT organizations shouldn't lock themselves into a hard disk drive-only strategy. Instead, they need to entertain the possibility of an SSD tier, or tier 0, for their most I/O-intensive servers.
No comments:
Post a Comment