Flash memory is taking the data center world by storm and creating disruptive opportunities to challenge the status quo, overcoming the capacity challenges of RAM and the performance challenges of hard disks to enable innovative new architectures.
This article discusses several different real-world use cases for SQL Server, MySQL web scale, and performance scale up within a turnkey appliance.
A Highly Available, Shared-nothing Microsoft SQL Server Powerhouse
When flash is placed within the server, as close to the CPU as possible (i.e. natively on the PCI Express bus), the media’s microsecond latencies enables databases to get the performance they need without having to go across the network to Tier 1 storage (via shared-nothing architectures).
But many organizations’ High Availability (HA) requirements have kept them tied them to back-end storage, limiting them to less effective architectures that rely on hard disk arrays, unnecessary legacy storage protocols, and millisecond-latency network transfers.
Enterprise software developers have been quick to realize flash’s potential and develop features to support HA without a SAN or NAS. A great example is Microsoft SQL Server 2012’s AlwaysOn, which automates server-based replication and failover for many different environments.
Online gaming platform company, BetOnSoft, used flash memory with AlwaysOn and readable mirror to achieve the following benefits:
- Eliminated maintenance job’s impact on performance
- 10.8x faster Check DB jobs
- 9x faster backups
- Support 10x more workload per server
- Offloads reporting with Fusion Powered readable mirror
- 200-millisecond average replication speeds (including network overhead)
- 14-second failover
Notably, Microsoft is not the only Enterprise vendor in the space to offer
these kinds of features. Oracle’s DataGuard provides similar server-based replication and failover—and flash memory eliminates performance slowdown for both.
Figure 1. BetOnSoft maintained HA and dramatically improved performance without a SAN
Web Scale Up, Not Out
MySQL (within a LAMP stack) is pervasive among web properties, such as eRetailers and social networking companies. The ability to simply add servers, implement memcache pools or RAM, etc., works very well in volatile environments that may have to scale quickly.
But as these companies begin to grow, performance per unit of scale out (servers) shrinks and operating costs grow. Flash memory solves this problem, maximizing the performance density of each server. It’s low latency and short queue depths also make the entire system more resilient to traffic spikes.
Drecom, Japan’s leading provider of social networking games for mobile devices, used flash memory to do just that. Drecom added flash memory to Master-Slave-Standby trios to achieve the following benefits:
- 20x lower I/O wait time
- 20x faster update queries
- Eliminated service interruptions to end-users, even under heavy traffic
- Growth headroom to support expanding product line
- 4:1 server consolidation
- Increase replication speeds to enhance reliability
Figure 2. Drecom consolidated MySQL scale out database footprint 4:1.
Scaling Up Performance without RAM
An interesting use of flash memory that is increasingly pervasive is as RAM replacement. Flash is much more affordable on a per GB basis than RAM and organizations can now place multiple terabytes of flash in servers. This makes it possible to fit even large data sets entirely into a high-performance (and even persistent) memory tier.
SevOne, a provider of turnkey enterprise network monitoring appliances put this capability to good use, enabling its customers to continue using a single appliance to achieve the following benefits:
- Real-time monitoring of millions of elements
- Dramatically faster real-time current and historical reporting over billions of baseline data points
“The [flash memory] performed like the data was in RAM. After testing them at our most challenging customer site, we decided to make them standard in two of our top-tier appliances,” says Tanya Shea, SVP of Operations, SevOne. Improved data protection by caching to non-volatile flash rather than RAM.
Flash memory (as opposed to flash storage, which is essentially a fast disk) offers exceptionally low-latencies to maximize end application throughput. Its high performance overcome the scale-out weakness of hard disk arrays, while its performance density often makes it a great fit to replace RAM. The examples described are just a sample of the many ways this flexible new media is changing data centers around the world.
About the Author
Gary has served in leadership roles at numerous data center infrastructure companies. Prior to Fusion-io, he was the vice president of marketing at MaxiScale, focused on web scale file systems and acquired by Overland Storage. Prior to MaxiScale, he was the vice
president of marketing and business development at Gear6, focusing on storage and web caching. He also served as vice president of marketing at Compellent which went public in 2007, and was a co-founder at Nishan Systems, which was acquired by McDATA/Brocade.