RAMcloud: Creating A New Class Of Storage
@smriti-ZtAJsx
•
Oct 26, 2024
Oct 26, 2024
1.4K
Researchers at Stanford University are working on the idea of making a RAM-based cloud, using DRAMs. DRAMs have been of extensive use in recent years accounting to the need for large-scale database systems. The target here is to provide primary storage for cloud computing environments. If the idea is successful, RAMcloud will implement projects which need large datasets much more intensively than what is available now.
#-Link-Snipped-#
Although DRAM as a storage device is intended for a RAMcloud, the limits of the same are well known to the Researchers. In majority cases, DRAM is only put to use as a cache for faster processing, in other cases, it works in a specialized way for an application-specific memory storage device. Noting this, the goal behind RAMcloud is to create a general-purpose system tackling the full potential of DRAM such that there are no cache misses. With the foundation of a durable RAMcloud, an alternative backing store wouldn't be needed, hence cancelling any backing store overheads that are associated with DRAM.
The project which was started in 2009, is still in its early stages and a lot of exploration is still to be done such as the introduction of secondary indexes and multi-object transactions. The network protocols are yet to be zeroed upon and performance has to be enhanced maintaining consistency. Decisions regrading Multiple Datacentres, Multi-Tenancy are still in the pipeline. The systems, as of now, are running for simple, small-scale tests.
Source: #-Link-Snipped-#
#-Link-Snipped-#
Although DRAM as a storage device is intended for a RAMcloud, the limits of the same are well known to the Researchers. In majority cases, DRAM is only put to use as a cache for faster processing, in other cases, it works in a specialized way for an application-specific memory storage device. Noting this, the goal behind RAMcloud is to create a general-purpose system tackling the full potential of DRAM such that there are no cache misses. With the foundation of a durable RAMcloud, an alternative backing store wouldn't be needed, hence cancelling any backing store overheads that are associated with DRAM.
The project which was started in 2009, is still in its early stages and a lot of exploration is still to be done such as the introduction of secondary indexes and multi-object transactions. The network protocols are yet to be zeroed upon and performance has to be enhanced maintaining consistency. Decisions regrading Multiple Datacentres, Multi-Tenancy are still in the pipeline. The systems, as of now, are running for simple, small-scale tests.
Source: #-Link-Snipped-#