Citation Context

... object stores. For example, AWS [8], Google Cloud [4] and OpenStack [10] provide their own Hadoop to object store connectors that allow tenants to directly use object stores as a replacement of HDFS =-=[26]-=-. Moreover, commercial Big Data platforms such as Amazon EMR [1] and Azure HDInsight [5] go a step further and directly employ object stores as the primary storage technology. Cloud-based object store...

Citation Context

...nants can control. We set the time slot for our models to one day. We assign to our workload, in a round-robin fashion, four analytics applications that are typical components of real-world analytics =-=[12, 29]-=- and exhibit diversified I/O and computation characteristics. Sort, Join and Grep are I/O-intensive applications. The execution time of Sort is dominated by the shuffle phase I/O. Grep spends most of ...

Citation Context

...ave traditionally been used as data dumps for large objects such as backup archives and largevolume pictures or videos; use cases where SSDs would incur a high acquisition as well as maintenance cost =-=[24]-=-, e.g., premature device replacement. Nevertheless, recent research has shown that SSDs can deliver significant benefits for many types of Big Data analytics workloads [13, 15, 16, 18], which are thus...

Citation Context

...s desirable for both cloud providers and tenants. To verify our argument, we conducted a trace-driven simulation study by replaying two 250-job snippet traces from Facebook’s Hadoop production traces =-=[12]-=-. We set the HDD tier price as $0.0011/GB/day—average of the Google Cloud Storage price of $0.00087/GB/day and the Google Cloud’s HDD persistent block storage price of $0.0013/GB/day—and the SSD tier ...

Citation Context

...trade-offs. For example, while the newer 3-bit MLC NAND technology promises to deliver higher SSD densities and potentially drive down the acquisition cost, it has taken a major toll on SSD endurance =-=[14, 23, 28]-=-, which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers [15,...

Citation Context

...as well as maintenance cost [24], e.g., premature device replacement. Nevertheless, recent research has shown that SSDs can deliver significant benefits for many types of Big Data analytics workloads =-=[13, 15, 16, 18]-=-, which are thus driving the need for adopting SSDs. Newer technology on this front is promising, but does not adequately address the cost and performance trade-offs. For example, while the newer 3-bi...

Citation Context

...ing insights into the advantages of dynamically priced tiered object storage management involving both cloud providers and tenants. Cloud Pricing Researchers have also looked at cloud dynamic pricing =-=[19, 21, 25]-=-. CRAG [3] focuses on solving the cloud resource allocation problems using game theoretical schemes, while Londono et al. [22] propose a cloud resource allocation framework using colocation game strat...

Citation Context

...as well as maintenance cost [24], e.g., premature device replacement. Nevertheless, recent research has shown that SSDs can deliver significant benefits for many types of Big Data analytics workloads =-=[13, 15, 16, 18]-=-, which are thus driving the need for adopting SSDs. Newer technology on this front is promising, but does not adequately address the cost and performance trade-offs. For example, while the newer 3-bi...

Citation Context

...ing insights into the advantages of dynamically priced tiered object storage management involving both cloud providers and tenants. Cloud Pricing Researchers have also looked at cloud dynamic pricing =-=[19, 21, 25]-=-. CRAG [3] focuses on solving the cloud resource allocation problems using game theoretical schemes, while Londono et al. [22] propose a cloud resource allocation framework using colocation game strat...

Citation Context

...trade-offs. For example, while the newer 3-bit MLC NAND technology promises to deliver higher SSD densities and potentially drive down the acquisition cost, it has taken a major toll on SSD endurance =-=[14, 23, 28]-=-, which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers [15,...

Citation Context

...trade-offs. For example, while the newer 3-bit MLC NAND technology promises to deliver higher SSD densities and potentially drive down the acquisition cost, it has taken a major toll on SSD endurance =-=[14, 23, 28]-=-, which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers [15,...

Citation Context

... 28], which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers =-=[15, 17, 20, 27]-=-. Data analytics applications are particularly amenable to such tiered storage deployments because of the inherent heterogeneity in workload I/O patterns. The choice of tiers depends on tenants’ workl...

Citation Context

...ing insights into the advantages of dynamically priced tiered object storage management involving both cloud providers and tenants. Cloud Pricing Researchers have also looked at cloud dynamic pricing =-=[19, 21, 25]-=-. CRAG [3] focuses on solving the cloud resource allocation problems using game theoretical schemes, while Londono et al. [22] propose a cloud resource allocation framework using colocation game strat...

Citation Context

... 28], which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers =-=[15, 17, 20, 27]-=-. Data analytics applications are particularly amenable to such tiered storage deployments because of the inherent heterogeneity in workload I/O patterns. The choice of tiers depends on tenants’ workl...

Citation Context

...as well as maintenance cost [24], e.g., premature device replacement. Nevertheless, recent research has shown that SSDs can deliver significant benefits for many types of Big Data analytics workloads =-=[13, 15, 16, 18]-=-, which are thus driving the need for adopting SSDs. Newer technology on this front is promising, but does not adequately address the cost and performance trade-offs. For example, while the newer 3-bi...

Citation Context

...ource allocation problems using game theoretical schemes, while Londono et al. [22] propose a cloud resource allocation framework using colocation game strategy with static pricing. Ben-Yehuda et al. =-=[11]-=- propose a game-theoretic market-driven bidding scheme for memory allocation in the cloud. We adopt a simplified game theoretic model where the cloud providers give incentives in the form of dynamic p...

Citation Context

...oud Pricing Researchers have also looked at cloud dynamic pricing [19, 21, 25]. CRAG [3] focuses on solving the cloud resource allocation problems using game theoretical schemes, while Londono et al. =-=[22]-=- propose a cloud resource allocation framework using colocation game strategy with static pricing. Ben-Yehuda et al. [11] propose a game-theoretic market-driven bidding scheme for memory allocation in...

Citation Context

...as well as maintenance cost [24], e.g., premature device replacement. Nevertheless, recent research has shown that SSDs can deliver significant benefits for many types of Big Data analytics workloads =-=[13, 15, 16, 18]-=-, which are thus driving the need for adopting SSDs. Newer technology on this front is promising, but does not adequately address the cost and performance trade-offs. For example, while the newer 3-bi...

Citation Context

... 28], which raises the maintenance costs. Tiered storage is used in many contexts to balance the HDD–SSD cost and benefits by distributing the workload on a hybrid medium consisting of multiple tiers =-=[15, 17, 20, 27]-=-. Data analytics applications are particularly amenable to such tiered storage deployments because of the inherent heterogeneity in workload I/O patterns. The choice of tiers depends on tenants’ workl...

Citation Context

...he advantages of dynamically priced tiered object storage management involving both cloud providers and tenants. Cloud Pricing Researchers have also looked at cloud dynamic pricing [19, 21, 25]. CRAG =-=[3]-=- focuses on solving the cloud resource allocation problems using game theoretical schemes, while Londono et al. [22] propose a cloud resource allocation framework using colocation game strategy with s...

Citation Context

... stores as the primary storage technology. Cloud-based object stores use low-cost HDDs as the underlying storage medium. This is because the price gap between HDDs and SSDs continue to be significant =-=[9]-=-, especially for datacenter-scale deployments. Object stores have traditionally been used as data dumps for large objects such as backup archives and largevolume pictures or videos; use cases where SS...

Citation Context

...undant data copying cost, cloud providers typically let their tenants run Big Data processing jobs on vast amount of data stored in object stores. For example, AWS [8], Google Cloud [4] and OpenStack =-=[10]-=- provide their own Hadoop to object store connectors that allow tenants to directly use object stores as a replacement of HDFS [26]. Moreover, commercial Big Data platforms such as Amazon EMR [1] and ...