The industry has long worried that cloud computing cannot deliver the performance required for critical enterprise applications. For example, the very notion of multiple compute instances or multiple applications sharing the same infrastructure has meant that service providers cannot guarantee service level agreements for response times. This has held many companies back from making the jump to the cloud. These performance bottlenecks are indicated in Figure 1. Multiple applications or virtual machines simply drive too much storage traffic to traditional disks, which cannot keep up with demands.

Figure 1: Multiple applications or compute instances drive storage and I/O performance bottlenecks

Moving data and applications to cloud infrastructures comes with a twofold challenge: maintain high system performance while data size continues to grow and constantly keep infrastructure expenses low.

Flash memory is able to address these two aspects: It increases performance density, enabling more applications or compute instances per server, shrinks the footprint and cost of data center hardware, and decreases latency to guarantee optimal response times.

Figure 2: Server-based NAND flash memory increases performance and decreases latency for cloud configurations

Let's explore the use of flash memory in several cloud architectures from web-scale computing to Software-as-a-Service to social gaming.

Web Scale
A large Internet services organization in Japan was facing problems scaling search for an ecommerce property. By relying on a disk-based infrastructure, it was facing huge upfront capital costs for marginal performance improvements, constant cost increases for space and power, and increasing system complexity, which would drive higher maintenance costs and decrease system reliability.

By implementing server-based NAND flash throughout the system, this organization was able to:

• Shrink the overall system footprint by 50%
• Slash power and cooling costs by replacing power-hungry disk arrays with NAND flash
• Lower maintenance costs and improve reliability with far fewer hard drives to maintain
• Immediately recognize an investment return by repurposing unused servers
• Achieve all of this at a cost $4 million dollars less than the closest competitive solution

The full details are shown in Figure 3.

Figure 3: Excessive infrastructure scale out was prevented by the adoption of server-based NAND flash technology

Software as a Service (SaaS)
Attracting customers to SaaS, another implementation of cloud computing, means giving them an end-user experience that performs effectively and predictably at an advantageous cost. A leading customer support SaaS provider faced the following needs as they tried to scale their MySQL infrastructure to support their application:

1. Guaranteeing top-tier, scalable performance, while improving the customer experience
2. Providing enterprise reliability to ensure mission-critical data was never lost and maintain 24/7/365 availability under any load
3. Remaining cost-effective, to ensure premium services at a competitive price

Sticking with a traditional disk-based infrastructure made it impossible for the organization to achieve these objectives. By implementing a PCIe-based NAND flash approach across six servers, they were able to solve their challenges, specifically they were able to:

• Enable end customers to create more complex ticket views
• Drive 2x faster MySQL index scans
• Increase 3x more data support without impacting performance
• Reduce warm-up time from six hours to three minutes
• Create a simpler, more reliable system
• Derive a lower total cost than competing solutions

Social Gaming
A more consumer-focused aspect of cloud computing is social gaming, which can experience millions of users descending on the infrastructure supporting the application. A large European hosting group was facing challenges with a browser-based MMOG (Massively Multiplayer Online Game) at one of its data centers.

The game provider maintained multiple virtualized databases, each with different I/O requirements. It had been scaling performance by continually adding servers, but the system's performance capabilities were reaching its limit. On top of this, the management burden and costs to maintain the growing server farm were beginning to strain the company's business model.

The hosting company was trying to maximize this game provider's infrastructure investment and flexibility by virtualizing the databases. However, the different I/O profiles resulted in unpredictable workloads that could not be consistently and affordably handled with traditional storage.

By using PCIe-based NAND flash memory products directly in the servers, these challenges were solved. What would have taken over 150 additional high performance hard disk drives across a 60-server hosting platform was reduced to just 24 servers with ample performance to run both applications and databases. Overall, the solution provided:

• Two to 5x more workload per server
• The ability to virtualize multiple databases with varying workloads
• Over 30% less hardware
• The ability to eliminate worry over application service interruptions
• The best price/performance ratio to extend more attractive pricing to customers

Figure 4: Dramatic consolidation by implementing server-based PCIe NAND flash memory

As these examples show, server-based flash memory can enable high performance cloud computing, even for data-intensive applications. When planning to deploy a cloud solution, be sure to ask vendors for examples like these to ensure that the prouduct you choose can deliver on its promises.