Industry’s Highest Performing I/O

I/O performance is critical for application throughput, user response time and CPU efficiency. It is a necessary part of a balanced system design. Without the required I/O performance, CPUs sit idle waiting for a response while users complain.

Amdahl postulated in the 1960’s that in a balanced system there is 1 byte of I/O for each instruction. This number has been reduced to 1 byte per 30 instructions through the use of caching and the changes in workloads. This would still imply that a quad socket — quad core server running at 2 GHz — would require 1 GByte/sec of I/O.  The typical server workload requires 8Kbyte accesses — or over 100,000 IOPS!

The figure below shows the performance of the Violin 3200 against alternate technologies over a wide range of block sizes. Most server workloads have an average I/O block size of between 4Kbytes and 16Kbytes. For these block sizes, the Violin 1010 provides 100x performance over standard hard drives and drive arrays.

The performance of VXM for Flash is significantly better than traditional Flash SSDs. With thousands of Flash devices in a single system, VMX supports algorithms that enable massively parallel operations. These algorithms enable very high bandwidth, lower latency and very high IOPS. In particular, sustained random WRITE IOPS over 100K can be supported.

Features

VXM is implemented in a high performance hardware architecture platform. Processors and operating systems are only used for management of a system. The data path has very low latency, high bandwidth and very high IOPS. Specific advantages of the VXM architecture:

  • Highest IOPS: The Input/Output Per Second (IOPS) of most storage systems is limited by the Hard Disk limits of around 3000 per drive. On a single Linux server, the Violin 1010 can support over 1 Million random or sequential IOPS, the equivalent of 3,000 hard drives.
  • Lowest latency: The Violin 1010 can connect directly to the PCI express I/O bus and avoid the latencies of HBAs, Fiber channel controllers and protocols. This enables a latency of 3 µsec, 1000x faster than a hard drive and 100x faster than Flash SSDs.
  • Highest bandwidth: The Violin 1010 can efficiently support 8 lanes of PCI express, a 20 Gbit/s interface.  This enables 1.7 GByte/sec of bandwidth from a 2U system. A pair of systems can support 3.4 GByte/sec to a single host.