Slow hosts frustrate users and hurt productivity. The good news? You have power over performance through optimization. The stakes are high to keep storage hosts running at top speed. Fortunately, with some adjustments, you can significantly optimize performance.
Many changes require only minor configuration tweaks rather than expensive hardware replacements. By implementing several of the tactics discussed here, you can transform even aging servers into high-speed workhorses. Your users will be amazed at how quickly everything runs.
Let’s dive deep and look at 10 tactics to optimize storage and boost speed.
Tactic 1: Upgrade to SSDs
One of the easiest ways to instantly speed up the storage performance of your data center is to switch from traditional hard disk drives to solid-state drives. SSDs have no moving parts, so they can access and write data much faster than HDDs. Make the change gradually by replacing your oldest HDDs with comparable-capacity SSDs as budgets allow. SSD performance will satisfy even your speediest applications.
Tactic 2: Adjust RAID Levels
RAID, or redundant array of independent disks, spreads data across multiple disks for redundancy and performance. But not all RAID levels are created equal. Higher RAID levels like 5 offer redundancy at the cost of speed due to parity calculations. For performance-critical servers, opt for RAID 0 or 10, which divide and mirror data without parity overhead. Match RAID levels to your needs, as the right choice can more than double sequential read-write speeds.
When adjusting your RAID levels, consider these factors:
- RAID 0 provides no redundancy but maximum speed by stripping data across disks. Best for temporary datasets or where backups are used.
- RAID 1 mirrors data to two disks for redundancy, doubling capacity as needed. Writes are slower than reads.
- RAID 5 distributes parity data across all disks, allowing any one disk to fail without data loss. Uses parity calculations, which impact write speeds.
- RAID 6 is similar to RAID 5, but can survive two simultaneous disk failures. Very robust but significantly slower than other levels due to double parity calculations.
- RAID 10 combines striping and mirroring for good performance and redundancy. Faster than RAID 5/6 for reads and writes but uses more disks.
- Consider disk characteristics like RPM, cache size, and interface type, which also impact I/O and suit different RAID levels.
- Benchmark your most demanding applications at different RAID levels to see true performance differences. Many will get big speed boosts from RAID 0 or 10.
The right RAID level is a balancing act of speed, capacity, cost and reliability for your data center workload needs. Take time to test and compare to get the optimal configuration.
Tactic 3: Tune Filesystems
Filesystem tuning is a digital tune-up that improves how your operating system organizes and accesses files. For Linux/UNIX hosts, adjust settings in /etc/fstab to mount filesystems for speed. Tweak flags like noatime, relatime and nodiratime reduce disk writes from metadata access. Ext4 usually outperforms EXT3, but test both. On Windows, defragment regularly and adjust the page file size. Taking the time for filesystem housekeeping pays off in streamlined performance.
Tactic 4: Install Latest Drivers and Firmware
Out-of-date drivers are like cars running on bald tires—they just can’t roll at top speeds. Make sure your servers have the latest storage drivers, RAID controller firmware, and chipset updates installed. Newer versions of software close security holes and optimize for current hardware. Spend an afternoon applying updates to every host. You’ll be shocked which ones were years out of date! Regular updates prevent slowdowns and unlock dormant speed potential.
Tactic 5: Use Appropriate I/O Schedulers
I/O schedulers determine how your storage devices will react to read and write requests. The default scheduler varies by OS but may not be best for your workload. Test alternate schedulers like deadline and cfq to see if they better suit demanding applications. Deadline prioritizes latency for responsiveness, while cfq balances latency and throughput. Benchmark different schedulers to find the right fit for your environment. A little scheduler tweaking can go a long way.
Tactic 6: Tune Memory Buffers
In Linux, the buffered I/O cache holds recently accessed disk data in RAM for quicker reuse. But its size defaults are often too low for performance-critical applications. Tune the disk read-ahead buffer, page cache, and other memory buffers up based on your hosts’ RAM. More memory means more data is cached, accelerating subsequent reads. On Windows, adjust the system page file to match physical memory for faster virtual memory paging. Crank up memory buffers for a memory-boosted boost.
Tactic 7: Investigate Filesystem Options
Filesystem creation options let you customize how data is stored. For databases or VMs with many small files, enable filesystem features like extents, which group allocations to reduce fragmentation. Test options like dir_index, filetype, and lazy_itable_init. Mount filesystems with commit=5 or commit=600 to tweak sync behavior. Play around to find the goldilocks configuration—not too slow and not too fast—that best suits each host role. Your efforts will be rewarded.
Tactic 8: Consider Non-Uniform Memory Access
In large NUMA node servers, data access speed depends on where files are located relative to CPUs. By default, Linux spreads files randomly, hurting performance. Use numactl to migrate files, assign memory, and set processor affinity for applications and filesystems to local NUMA nodes. This reduces cross-node memory hops for faster access. On Windows, enable network location awareness and preferred NUMA node support. NUMA optimizations align workloads to fly at top speeds.
Tactic 9: Tune Network Buffer Settings
For high-performance NFS and iSCSI, network buffer tuning is critical but often neglected. Linux offers many knobs to play with, like enqueuing, completing, dirty page writing, read-ahead windows, and request/response timeouts. Tweak settings conservatively, then benchmark for optimal values. On Windows, adjust the TCP window size and receive/send buffers. With the right network configuration, your storage traffic will zip across the fastest lanes.
Tactic 10: Add Drive or Server Caching
Caching is the ultimate hack for turbocharging storage I/O. At the device level, enable SSD read/write caches. For servers, setup DRAM, SSD, or PCIe flash caching using solutions like bCache or FlashCache. Caching intercepts and satisfies common requests from ultra-fast media versus spinning disks. With a smart caching layer, even gigantic datasets feel zippy. Plus, it extends the usable lifespan of your SSDs.
Final Words
With these 10 tactics in your toolbox, your storage performance will be through the roof. Don’t be daunted—start with one change at a time. Monitor before and after with tools like iostat and sar. Fine-tune adjustments based on how workloads perform. A little optimization work now saves major headaches later and ensures your data center remains in the fast lane. Soon, servers will fly, and users won’t know how you did it.
