Memory Motion
Memory Motion brings RAM resources to the servers that need it most
Memory Motion is an MVX capability that brings RAM resources to the servers that need it most. MVX consolidates RAM resources from hosts around the data center into a centrally-managed pool, and brings it to guest machines that could use more RAM. The guest machines benefit by having a large and fast swap space, so their operating system can address a much larger memory space efficiently.
Making swapping work for you
Every operating system and hypervisor uses a swap device to manage its memory. The swap is used most intensively when system load is high. Unfortunately, this is precisely the time when I/O delays cause significant performance degradation. Application speed can become intolerably inconsistent, falling precipitously when the operating system starts swapping. CPU utilization often falls as well, as the system waits for its swap device to deliver memory. Up to now, most data centers simply try to eliminate swapping by adding more RAM to every system, even though the RAM will be poorly utilized. Even then, some servers have peak needs beyond their physical RAM, and older machines may already run at their maximum memory limits.
Chart: Dataset Size to Memory Ratio
With MVX 2.5, swap space can be implemented in the MVX Memory Cloud, enabling swapping to shared RAM rather than to disk. Analysis has shown that the MVX Memory Cloud can sustain swapping at sustained rates of over 700 megabytes per second, with even greater speeds if multiple swap devices are allocated. This ultra-fast swapping makes system performance much more stable as load increases. It also leads to higher CPU utilization by drastically cutting swap wait times. This can be particularly valuable for hypervisors, for which memory management is a challenging problem, and disk-based swapping can destroy performance.
In one data center, the impact was dramatic. An application was tested on a server with no swapping, and then on the same server with less and less RAM. Compared to local disk swapping, and even to solid-state device (SSD) swapping, RNA MVX cut wait times by 26X.
Bringing new levels of performance to your data center
Through Memory Motion, new system configurations become possible. Today, CPUs come with 64-bit memory addressing, but a server operating system will only utilize a small fraction of this address space. A server OS will run within a small multiple of its physical RAM, to avoid the penalty of too much swapping. With MVX, a server can have an ultra-fast swap device with a terabyte or more of shared RAM behind it. The operating system can fill that entire address space with data so that number-crunching processes can run without disk delays.
Memory Motion can also allow a reallocation of memory across clusters. Once every system can access a fast swap space, servers can be provisioned with less RAM, reducing equipment and power costs. Alternatively, more server RAM can be dedicated to application processes to yield faster and more consistent application performance. In effect, Memory Motion enables RAM load-balancing across an application cluster, or even across the data center.
These benefits are all possible without risky changes to application programs or hardware. Because MVX is simply mounted like a block device for Memory Motion, there is no application integration. And because it interfaces with operating systems and hypervisors as swap space, it can be applied to performance issues anywhere in the data center.
