Data-in-flash beats Data-in-DRAM by up to 17X writes Wikibon CTO David Floyer in his latest report on leading-edge storage technology, “Data in DRAM is a Flash in the Pan”. This, he says, means that medium-to-large organizations and especially Cloud service providers should avoid DRAM-based solid-state solutions such as Cassandra and SAP’s HANA. Instead they should implement flash-in-server systems such as Aerospike, which he names as the current leading system for transactional analytic applications, or FusionIO. He also recommends that major vendors such as IBM, Oracle, and Teradata radically change their architectures to embrace a true data-in-flash model.
The recommendations are based on detailed analysis of the price, performance, and cost of operation of the two technologies. The central finding is that the total two-year data-in-memory server costs for a 2 Tbyte replicated transactional analytic database with 5,000 bytes per operation using data-in-flash is $73,000, versus a total cost for a data-in-DRAM solution of $1,209,000, 17 times higher.
The main reason for this difference is that data-in-DRAM requires 50 nodes to meet the performance requirements, while flash-in-memory requires only four. Furthermore, the cost of the DRAM solution increase steadily as the number of bytes-per-operation grows from 100 to 10,000, while the cost of flash remains steady.
The report also looks at the advantages of the enhanced and lower overhead methods of addressing flash memory announced recently by Fusion-io, while noting that other vendors are working on similar extensions, and projects the likely development path of flash-in-memory systems. Floyer concludes that this will only increase the cost differential between the flash and DRAM approaches over time.
Since the cost advantage of flash grows steadily with the increase in the amount of data the system handles, the Big Data revolution will also drive the price/performance advantage of flash over time, making DRAM systems a competitive disadvantage for users. This is not just an issue for large enterprises and Cloud service providers, although they will feel the differential most acutely in many cases. Floyer cites his case study “The Hunting of the RARC” about the experience of a traditional mid-sized company that achieved impressive operational savings by implementing a data-in-flash system for its ERP system as an indication of the importance of this issue for mid-range companies.
As a result, he writes, “Wikibon would strongly recommend that executives from ISVs and enterprises support a data-in-flash model, especially for high performance and Big Data systems. Data-in-DRAM should be bypassed.”
As with all Wikibon research, this report is publicly available in its entirety without charge on the Wikibon Web site. Interested IT professionals should register for free membership in the Wikibon community, which allows them to comment on research, post their own research and questions for the Wikibon analysts, and participate in the frequent Peer Incite meetings, at which IT professionals discuss their experiences solving real-world business problems.
Latest posts by Bert Latamore (see all)
- Wikibon identifies ‘galaxies’ forming from big data ‘big bang’ - July 27, 2016
- Adoption complexity is holding back big data growth, says Wikibon - July 26, 2016
- Wikibon says It’s too early for Big Data performance benchmarks - July 25, 2016