Dictionary sharing: An efficient cache compression scheme for compressed caches

Abstract

The effectiveness of a compressed cache depends on three features: i) the compression scheme, ii) the compaction scheme, and iii) the cache layout of the compressed cache. Skewed compressed cache (SCC) and yet another compressed cache (YACC) are two recently proposed compressed cache layouts that feature minimal storage and latency overheads, while achieving comparable performance over more complex compressed cache layouts. Both SCC and YACC use compression techniques to compress individual cache blocks, and then a compaction technique to compact multiple contiguous compressed blocks into a single data entry. The primary attribute used by these techniques for compaction is the compression factor of the cache blocks, and in this process, they waste cache space. In this paper, we propose dictionary sharing (DISH), a dictionary based cache compression scheme that reduces this wastage. DISH compresses a cache block by keeping in mind that the block is a potential candidate for the compaction process. DISH encodes a cache block with a dictionary that stores the distinct 4byte chunks of a cache block and the dictionary is shared among multiple neighboring cache blocks. The simple encoding scheme of DISH also provides a single cycle decompression latency and it does not change the cache layout of compressed caches. Compressed cache layouts that use DISH outperforms the compression schemes, such as BDI and CPACK+Z, in terms of compression ratio (from 1.7X to 2.3X), system performance (from 7.2% to 14.3%), and energy efficiency (from 6% to 16%).

Safecracker: Leaking Secrets through Compressed Caches

“…3.1). As previously discussed, we choose VSC because it is the most commonly used in cache compression work [8,9,17,41,42,60,70,81]. There are two differences between VSC and our simple example.…”

“…In this paper, we provide an answer in the affirmative by analyzing the security of memory hierarchy compression, specifically cache compression. Compression is an attractive technique to improve memory performance, and has received intense development from both academia [3,4,8,9,37,40,55,59,60,62,69,70,81,90,92] and industry [17,18,31,41,45,61]. Several deployed systems already use memory-hierarchy compression.…”

Dice

“…Compression exploits redundancy in data values to increase effective capacity of a given substrate. Prior work has looked at compression for improving the capacity of SRAM caches [4,28,34,35]. As decompression latency is in the critical path of cache access, these proposals use simple compression schemes such as Frequent Pattern Compression (FPC) [5], Base-Delta-Immediate (BDI) [31], CPACK [11], and ZCA [17], that can perform decompression within a minimal number of cycles.…”

“…Therefore, LCP can utilize compression not only for capacity but also for bandwidth benefits. Unfortunately, the page mapping and organization of LCP must be done using the OS, as the OS is required to know the compressed page size, in order to access the adjusted 1 Recent studies on SRAM cache compression propose sharing tags between a larger number of sets (say 4x sets, called superblocks) to reduce tag and metadata overhead [28,34,35]. However, when applied on DRAM caches, these designs increase the number of sets that must be checked on each access, which can waste bandwidth.…”

“…As such, 11.4% of our 19.0% speedup is from bandwidth benefits, not capacity (DICE over TSI). Recent proposals, such as Skewed Compressed Cache (SCC), investigate reducing SRAM tag overhead by sharing tags across spatially-contiguous sets in what are called superblocks [27,28,34,35]. For a 4x-superblock 8-way physical cache, these proposals use 32 physical tags to address up to 128 compressed lines by sharing the tags of neighboring sets.…”