We believe that selecting the right technology can be reduced to understanding and managing tradeoffs. Just as you might not purchase a Ferrari for family outings, you’d think twice about purchasing spinning disk for IOPS, Flash for latency-sensitive consistency, or DRAM for bulk storage.
Current performance I/O commonly relies on storage built around Flash and spinning disk, frequently buffered with DRAM. Kove memory meshes provide an altogether different performance platform, providing nanosecond consistency right into the CPU.
Advantages. Spinning disk continues to offer advantages for bulk storage. It can parallelize sufficiently enough to generate high bandwidth that does not require low latency. Spinning disk is reliable, proven, and inexpensive. Disk arrays do and will bulk up and failover reliably in the real world.
Disadvantages. Spinning disk has very poor latency and I/O Operations per Second (IOPS). Delivering high performance with spinning disk requires substantial power.
Advantages. Compared to spinning disk, Flash offers superior IOPS, read performance, and IOPS cost. Compared to DRAM, Flash provides superior energy efficiency and doesn’t require a backing store for long-term persistence.
Disadvantages. Compared to spinning disk, Flash is complex to manage. Flash performance can vary according to I/O pattern (read, write, mixed, burst, continuous, random, sequential), and its performance level can degrade as much as 90+% as its occupancy increases (“how full the disk is”). When performance and predictability matter, administrators monitor occupancy. Flash uses buffering to conceal 1.4 ms write costs. Constant, durable load can exceed buffering and result in poor performance.
Compared to DRAM, Flash performance varies considerably across I/O profiles, depending on whether applications are reading or writing, using small or large block sizes, using sequential or random access patterns, or running continuously or in burst. Usage wear influences performance, and wear leveling, block retirement, and RAM buffering diminish, but do not eliminate, this concern. Even with buffering, Flash latency lags behind DRAM performance as soon as buffers fill.
Latency, bandwidth, and IOPS all degrade through time in Flash. Performance varies across block size, and administrators should test each manufacturer’s product against their application to ensure that manufacturer design optimizations actually deliver against requirements. Flash can provide cost-effective burst performance, but does not yield consistent, sustainable performance under load over time.
Advantages. DRAM will run any I/O pattern at predictable performance levels, day-in and day-out, just as well for one year as for one minute, which is why it is used for the computation inside the CPU..
DRAM continues to provide the highest performing media, as well as the most robust solid state data access medium.
Compared to spinning disk or Flash, DRAM provides predictable, uniform, and superior performance for:
DRAM delivers speed without performance variability, durability, and dependability challenges, or reduced or inconsistent latency, bandwidth, and IOPS.
Disadvantages. Compared to spinning disk or Flash, DRAM costs more on a capacity basis. DRAM requires power backup and backing store for durable persistence across power cycles. DRAM uses less power on a performance basis, but more power on a capacity basis than either spinning disk or Flash.
When performance requirements matter to your bottom line, DRAM delivers:
DRAM costs more on a capacity basis, but can be less expensive on a performance basis.
Depending on application requirements, spinning disk provides economical and robust bulk storage. Flash provides burst performance, energy savings, and superior performance to spinning disk. DRAM provides the most robust approach to achieve the highest available performance across diverse and variable application requirements.
Kove® DRAM memory meshes provide greater versatility than storage, and provide nanosecond consistency right into the CPU.