Web 2.0

E-commerce and social networking has exploded in the past few years. Huge e-retailers such as Amazon.com, social networks like Facebook, and Software-as-a-Service vendors like Salesforce.com provide large scale and complex online applications. These applications are updated to add features as frequently as every week. Between traffic growth and application development, system load scales up constantly. Meanwhile, users expect consistently fast site performance; slower sites lose customers.

To meet the challenges of scalability and performance, web 2.0 system managers employ distributed computing with horizontal scaling. Web requests are load-balanced across arrays of commodity servers running common software and working from a common set of data. This provides the raw volume of compute capacity, but it opens a sharing and scaling problem. For example, when a request comes into a load-balanced server farm, data about logged-in users and their ongoing transactions must be shared.

With thousands of simultaneous users and hundreds of active servers, the sharing of and rapid access to this data becomes a major computing problem. Storage networks and data base servers offer one solution, but they impose a speed penalty, and they have hard limits on their capacity. Caches in RAM or solid state drives (SSD) in each server offer some relief, but they are limited by the size of each cache, the need to duplicate data into each server, and the large volume of data that may be accessed by any user. Software frameworks allow for some data objects to be shared among servers, but they are complex and require careful coding and management.

Memory Virtualization with MVX offers a breakthrough solution to the central scaling problems of web 2.0 services.

MVX breaks the bottleneck

MVX private cloud technology creates a large distributed memory service
- a Memory Cloud - from RAM in existing servers

Horizontal scaling works well for distributing compute load among web 2.0 servers and growing it by adding more servers. RNA MVX creates a Memory Cloud in a similar horizontal fashion, aggregating DRAM memory resources across servers in an elastic, expandable way. Each Memory Guest server can access the over the data center network.

The value for the web 2.0 enterprise starts with the size and sharability of RNA’s Memory Cloud. For disk-resident data objects, the MVX Memory Cache capability provides a shared file cache that can grown to a terabyte or more in size. For many web 2.0 services, this allows essentially all NAS-based file content to be served at RAM speeds while eliminating file load on their storage servers. By sharing RAM among all the application servers, the cache is far larger than any one server could otherwise contain, without purchasing additional hardware. Furthermore, the cache grows as new members are added to the server farm and they join the Memory Cloud, contributing some of their RAM and network bandwidth to expand the effectiveness of the common cache.

MVX provides a flexible and fast shared memory system for active data
regardless of size or quantity

MVX can also improve scalability in a second way. Memory Store capabilities provide an extremely flexible and fast shared memory system for large active data regardless of size or quantity. Like Memory Cache, Memory Store economizes on memory by keeping a single copy of the data rather than keeping multiple copies across the server farm; and it scales to large numbers of compute nodes. Unlike Memory Cache, Memory Store works without a using a NAS servers or NAS protocol, cutting overhead and making it a good fit for fast-changing session and index data.

Finally, the combination of these tools can open opportunities for improved services from web 2.0 server farms. One example concerns log file analysis. Memory Cache can be used to retain usage logs on one or more NAS servers while keeping the most recent data in the central Memory Cloud. Analytic software can run on any computer in the data center to access those logs far faster via Memory Cache, without impacting the performance of the web servers themselves. This can make real-time analytics practical where it was not before.

Proven results

MVX has proven its effectiveness in web 2.0 settings. One large on-line service improved query speed 100X, while increasing simultaneous user capacity 4X. Another replaced its Java caching systems with RNA MVX for keeping large session results for each user in the Memory Cloud. They built a multi-terabyte memory pool from the DRAM already in their server farm, and reduced the complexity of their system by removing the software caching systems. The result was an expansion of capacity from 100,000 to over 1 million simultaneous user sessions, without additional hardware.

RNA MVX provides a fundamental improvement in data center architecture by adding a shared memory system within the compute tier. RNA MVX gets more value from poorly utilized RAM and network capacity, thereby breaking storage bottlenecks, raising CPU utilization, and cutting user response time. As a systems technology using standard block and file I/O, it simplifies data management and frees software developers to concentrate on new functionality rather than scaling problems. The end results are happier users and more online revenue captured every day.