Replicate and Bundle (RnB)
Kalyani Khandezod, Nitin Raut, Abdulla Shaik · 2014
------------------------------------------------------------ABSTRACT----------------------------------------------------This work addresses the measurability and potency of RAM-based storage systems whereby multiple objects should be retrieved per user request. Here, a lot of the central processing unit work is per server dealing, not per requested item. Adding servers and spreading the information across them additionally spreads any given set of requested things across additional servers, thereby increasing the full variety of server transactions per user request. The ensuing poor measurability, dubbed the Multi-get Hole, has been reported in net a pair of.0 systems mistreatment memcached – a well-liked memory-based key-value storage system. We tend to gift Replicate and Bundle (RnB), a somewhat unintuitive approach: instead of add CPUs, we tend to add memory. Object replicas area unit mapped “randomly” to servers, and requested objects area unit bundled, choosing replicas therefore on minimize the quantity of servers accessed per user request and therefore the full central processing unit work per request. We tend to studied RnB via simulation within the context of DRAM-based storage, utilizing small benchmarks and enforced RnB modules for standardization. Our results show that RnB considerably reduces the quantity of transactions per request, creating operation additional economical. Also, in contrast to most alternatives, RnB permits versatile growth and comparatively straightforward readying. Finally, in systems whereby knowledge is replicated for different reasons, RnB is sort of free.