Decoupled zero-compressed memory

Abstract

For each computer system generation, there are always applications or workloads for which the main memory size is the major limitation. On the other hand, in many cases, one could free a very significant portion of the memory space by storing data in a compressed form. Therefore, a hardware compressed memory is an attractive way to artificially increase the amount of data accessible in a reasonable delay.Among the data that are highly compressible are null data blocks. Previous work has shown that, on many applications null blocks represent a significant fraction of the working set resident in main memory. We propose to leverage this property through the use of a hardware compressed memory that only targets null data blocks, the decoupled zero-compressed memory. Borrowing ideas from the decoupled sectored cache [12] and the zero-content augmented cache [7], the decoupled zero-compressed memory, or DZC memory, manages the main memory as a decoupled sectored set-associative cache where null blocks are only represented by a validity bit.Our experiments show that for many applications, the DZC memory allows to artificially enlarge the main memory, i.e. it reduces the effective physical memory size needed to accommodate the working set of an application without excessive page swapping.Moreover, the DZC memory can be associated with a zero-content augmented cache to manage null blocks across the whole memory hierarchy. On some applications, such a management significantly decreases the memory traffic and therefore can significantly improve performance.

A Survey Of Architectural Approaches for Data Compression in Cache and Main Memory Systems

“…Afterwards, the most probable elements are coded with [8, 10, 12, 16, 17, 23, 28, 29, 35, 40, 48-50, 52, 53, 56, 63, 64, 66, 78, 82, 83, 88, 93] Zero-content compression [14,35,41,50,52,57,62,78,81] Frequent value compression (FVC) [35,52,59,60,67,70] Base delta immediate (BDI) compression [16,23,35,48,51,63,83,93,94] LZ and variants [6,7,9,17,19,20,22 [8,12] Compression in GPUs [89,93,94] Use of compiler [9,21,43,44,72,73,93] fewer number of bits than those which are least probable. Thus, Huffman coding uses a variablelength coding scheme.…”

“…narrow-width data. Dusser et al [81] apply null data compression in main memory and also show that by combining their design with ZCcache design provides an opportunity to manage null blocks throughout the whole memory hierarchy.…”

Skewed Compressed Caches

“…Ekman et al [23] propose a compressed memory design that extends page table entries to keep compressed block sizes. The decoupled zero-compressed memory [19] detects null blocks using a design based on decoupled sectored set-associative caches. Pekhimenko, et al [14] propose a new compressed memory design that exploits compression locality to simplify address mapping.…”

“…Similarly, IIC-C [13] supports variable compressed blocks, but uses forward pointers to associate a tag with its sub-blocks resulting in high cache area overhead (24%). Dusser et al [19] augment the conventional cache with a specialized cache to store zero blocks (ZCA). A recent compressed cache design, DCC [26], tracks super-blocks to track more blocks with low area overhead.…”

Hicamp

“…Memory Compression has been proposed for main memory [14], caches [4,12] or both [13,19]. A mergeable cache architecture was designed [8] to de-duplicate the content of second level cache.…”