188宝金博页面版

Thrifty: Offering Parallel Database as a Serviceusing the Shared-Process ApproachPetrie Wong †‡∗ Zhian He ‡ Ziqiang Feng ‡ Wenjian Xu ‡ Eric Lo ‡† Department of Computer Science, The University of Hong Kong‡ Department of Computing, The Hong Kong Polytechnic Universitykfwong2@cs.hku.hk {cszahe, cszqfeng, cswxu, ericlo}@comp.polyu.edu.hkABSTRACTRecently, Amazon has announced Redshift, a Parallel-Database-as-a-Service (PDaaS). Redshift adopts the “virtual cluster”approach to implement multi-tenancy, which has the meritof hard isolation among tenants (i.e., tenants do not inter-fere even when sharing resources). However, that benef i tcomes with poor resource utilization due to the signif i cantredundancy incurred in the resources.In this demonstration, we present Thrifty, a Parallel-Database-as-a-Service operated using the “shared-process” approach.Compared with Redshift, each tenant in Thrifty does not oc-cupy an exclusive amount of resource but share the databaseprocesses together, leading to better resource utilization. Toavoid contention among tenants, Thrifty uses a proper clus-ter design, a tenant placement scheme, and a query routingmechanism to achieve soft isolation.In the demonstration, an attendee will be invited to reg-ister with Thrifty as a tenant to rent a parallel databaseinstance. Then the attendee will be allowed to view thedashboard of a Thrifty’s administrator. Next, the attendeewill be invited to control (e.g., increase) the workload ofthe tenant so as to see how Thrifty carries out online re-consolidation and elastic scaling.1. INTRODUCTIONIn this demonstration, we present Thrifty [11], a proto-type Parallel-Database-as-a-Service (PDaaS) operated usingthe shared-process approach. The shared-process approachadvocates the sharing of database installation and processamong tenants [4]. Thrifty adopts the shared-process ap-proach and utilizes the fact that real multi-tenancy environ-ments have active tenant ratios as low as 10% [8], so thatit can consolidate tenants that are active in dif f erent peri-ods on to the set of machine nodes running a single paralleldatabase process. Thrifty aims to deal with a general and∗ This work was done while af f i liated with ‡; This work ispartly supported by the Research Grants Council of HongKong (GRF PolyU 520413 and 521012) and a research giftfrom Microsoft Hong Kong.Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor prof i t or commercial advantage and that copies bear this notice and the full cita-tion on the f i rst page. Copyrights for components of this work owned by others thanACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-publish, to post on servers or to redistribute to lists, requires prior specif i c permissionand/or a fee. Request permissions from permissions@acm.org.SIGMOD’15, May 31–June 4, 2015, Melbourne, Victoria, Australia.Copyrightc ?2015 ACM 978-1-4503-2758-9/15/05 ...$15.00.http://dx.doi.org/10.1145/2723372.2735352.realistic setting of providing a high availability PDaaS tothousands of tenants. Each tenant requests a multi-nodeparallel database that serves multiple users. Queries mightbe submitted by a report generating application in a batch,or submitted by a user sequentially in an interactive analysissession. The query sets may be known or unknown before-hand. For report generating applications, the query tem-plates are known and could be found in the applications’stored procedures. For interactive analysis, however, a dataanalyst may craft and submit an arbitrary/ad-hoc query atany time. Thrifty is able to deal with that general settingusing a simple methodology called Tenant-Driven Design(TDD), which consists of three parts: (a) [Cluster Design]arrange the machine nodes in the cluster into groups and cre-ate a parallel database for each group of nodes, (b) [TenantPlacement] assign each tenant to several parallel databases(for replication), and (c) [Query Routing] route a tenant’squery to a parallel database containing that tenant’s dataat run-time.Our preliminary results [11] show that in a setting of 5000PDaaS tenants, where each tenant requests 2 to 32 nodesparallel database to query against 200GB to 3.2TB of data,Thrifty can serve all the tenants with a 99.9% performanceSLA guarantee and a high availability replication factor of3, using only 18.7% of the nodes requested by the tenants.The full technical details of Thrifty can be found in [11].2. THRIFTY OVERVIEWSystem AdministratorTenantsQuery ?Log……Shutdown Node n 406 -n 600DB Node n 1 -n 8 Tenant T 1 -T 10DB Node n 401 -n 406 Tenant T 100 -T 110……Deployment PlanTuningDeployment AdvisorTenant Activity MonitorQuery RouterDeployment MasterDeployment PlanQuery LogReplication Factor RQueryQueryThriftySLA Guarantee PTenant Activity………DB 0……DB 1 DB 100……DB Node n 71 -n 77 Tenant T 55 -T 70…………Figure 1: The architecture of the Thrifty system.1063