A New Kind Of Computer Memory Technique That Boosts App Speeds

A team of electrical and computer engineers recently developed a new technique that permits the computer processors to extract data with greater efficiency. With the inculcation of this new technology the application speeds will improve by 9.5 percent while the power consumption will come down by 4.3 percent on an average. The new technique introduced by these engineers is known as Dense Footprint Cache (DFC) which is actually a memory design. It brings great improvement in present day designs with the help of Dynamic Random Access Memory (DRAM) caches. These are mostly used for storage of frequency accessed data that comes close to computer processor. This implies that the processor wouldn’t need to waste time in deriving data from off-chip main memory. 

 

 

The cache data is usually organized in form of two huge blocks, the Mblocks. These helps the processor in finding the data it actually needs, But these Mblocks often come with multiple bytes of data that is irrelevant with the processor, therefore, their extraction wastes a lot of energy and time. The present day methodologies used to solve the problem by leaving “holes” in Mblocks that act as wasted space for what could be used in storage of useful data. DFC easily takes of both these problems. It mainly depends on useful block prediction to know which kind of data the processor will actually need from these Mblocks. Thus, the cache is able to do two things, one, compresses the Mblock for retrieving only required information and two, frees up space in cache for other kind of data that processor will mostly need. 

 

Another major benefit of DFC is that it brings down the frequency at which last-level cache misses out. This happens when processor tries to retrieve cache data that is actually not there. The misses compel the processor for extraction of data from main memory decreasing efficiency of the system. DFC reduces the LLC misses ratio by approx. 43 percent. The team working on this project ran simulation of some Big Data applications that could show the DFC results. They also introduced multiple solutions  for the challenges associated with this new concept, like placement and replacement of policies.