Memory Store

Many applications perform temporary or intermediate I/O as they run. They may generate intermediate results for many purposes: data analysis and modelling; what-if calculations or data warehouse reporting; graphic format conversions; or temporary indexing during sorting and data retrieval. This temporary I/O usually goes to disk, either because the data are too large for one server’s RAM, or because the application is designed to leverage its operating systems’ block and file I/O services. Under heavy load, however, the limited speed of disk I/O slows application performance, and the physical limitations of the disk impose a hard limit on the capacity of the application.

The Memory Store capability of RNA MVX overcomes these limitations. MVX provides any number of “virtual RAMdisks” in the Memory Cloud. Any server in the data center can mount an MVX block storage device directly, to get I/O at blazingly fast speeds. In addition, a single MVX block device can be mounted by more than one server when appropriate, enabling extremely fast communication between servers as well. Here’s how the Memory Store works.

Faster local computation

MVX Block Device
Block Device Using MVX Memory Store
MVX Shared Block Device
Shared Block Device using MVX Memory Store

MVX provides any server a block device in the Memory Cloud, which can be any size, up to the total RAM in the RNA Memory Cloud. Any application or database that does block I/O can use the MVX block storage device directly, just like a locally attached disk drive. If an application needs a standard Linux file system, such as ext2 or ext3, it is also supported; it is simply configured to use the MVX block device as its disk.

File I/O to the RAM-based Memory Store runs at speeds far greater than disk, limited only by CPU and network speed, for a very high rate of reads and writes to the application. Being RAM-based, the throughput provided by the MVX Memory Cloud is far greater than SATA disks or SSD can provide.

Faster distributed computation

The benefits of MVX Memory Store are not limited to single systems. MVX block devices can be made available to all memory guests, and shared through a cluster file system. The clustered file system ensures the integrity of all shared data. It can be configured simply, using the MVX block device as a shared disk system. The result is fast I/O, and fast inter-system communications.

Distributed application servers can use a shared Memory Store gain fast access to common datasets. This particularly benefits distributed analytics and modelling, which often reference a common model or data map. These reference data sets may occupy dozens or hundreds of gigabytes, making them too large to stay in any single server’s RAM. When placed in the RNA Memory Cloud, however, they can be shared among a number of servers while providing unmatched speed.

In addition, results from one guest server can be handed off to another server through the Memory Cloud, simply and without limits on the size of the results. This can greatly reduce the latency from server handoffs in distributed systems.

Flexibility and visibility across your data center

MVX Chart: Block Device Throughput

Beyond speed, MVX provides management advantages as well. Servers are often upgraded by adding RAM, and applications may be improved by configuring local RAMdisks. But the RAM benefits computation on only a single machine, and a conventional RAMdisk is limited in size to its server’s local RAM capacity. RNA’s Memory Cloud provides memory as a service to the entire data center, yielding flexibility benefits in many settings:

  • servers that have maxed out their RAM capacity can keep growing
  • servers redeployed into analytics farms may participate without RAM increases
  • virtual machines can employ additional RAM even when their hosts are limited
  • legacy applications that have hit a bottleneck can run faster without recoding

The RNA Memory Cloud provides a single cloud of RAM across a data center, with a single point of configuration and visibility, where managers can evaluate how memory resources are being used. The MVX Memory Store adds an additional kind of visibility. Since the MVX block device can be mounted from more than one server, a software developer or systems administrator can mount the device for inspection and debugging, without disturbing the running application.

Speed, flexibility, and visibility: They come together in RNA MVX.