OBJECTIVITY/DB – FLEXIBLE DEPLOYMENT OPTIONS

Leon Guzenda

Sujets

• Linternaute.com
• Réseau de stockage SAN
• Livre blanc
• BWA
• Cluster
• Circuit imprimé

Informations

• Publié par: Ulla
• Nombre de lectures: 90
• Langue: English

Extrait

An Objectivity and Panasas White Paper

Panasas ActiveStor Benchmark

THE BENEFITS OFAN OBJECTIVITY/DB AND PANASAS PARALLEL STORAGE HIGH-THROUGHPUT SOLUTION: A BENCHMARK REPORT

Background

Objectivity/DB uses a powerful, distributed processing architecture to manage localized, centralized or distributed object databases. The Panasas® ActivStor Parallel Storage Cluster accelerates time to results with highly reliable, massively scalable storage solutions for performance-driven applications in Government R&D, Oil & Gas, Manufacturing, Higher Education & Research, Life Sciences, CAE manufacturing, and emerging financial markets.

This white paper describes a benchmark that compares the performance of Objectivity/DB running against a local file system versus using files stored on a Panasas ActiveStor configuration. It covers the main features of Objectivity/DB and the ActiveStor equipment before explaining the benchmark software and the hardware configuration. It then presents and interprets the benchmark results. The benchmark demonstrated significant write speed and dramatic read speed improvements over conventional disk storage.

Objectivity/DB Overview

Objectivity, Inc.is the leading provider of distributed, real-time, SOA-enabled object database management solutions for mission-critical applications.The company’s flagship product, Objectivity/DB, is primarily used in Extremely Large Databases in the DoD, DHS and DoE communities and in complex real-time process control and telecom hardware and systems.

No Mapping Layer -Objects are stored directly to disk without translation into the rigid column and row format of relational databases. This avoids the expensive and inefficient mapping layers needed between object oriented languages, such as C++, C# and Java, and the database storage mechanism.Objectivity/DB is extremely fast at navigating complex networks and tree structures as there is no relational join table processing.

The main architectural featuresof Objectivity/DB are: ·The “single logical view” of distributed databases. ·Simple, distributed server architecture. ·Distributed Parallel Query Engine with replaceable components. ·Predictable scalability

Copyright © Objectivity & Panasas 2008

1 of 7

An Objectivity and Panasas White Paper

Panasas ActiveStor Benchmark

Single Logical View- Users see a single Federated Database that consists of local or distributed databases. The databases contain application controllable clusters of objects, relationships, collections and indices.

Simple, Distributed Server Architecture -Objectivity/DB is directly linked with client applications. It accesses simple, distributed servers that manage transaction management, remote data access and fragments of parallel queries. It can be configured in standalone, networked, peer-to-peer, Service Oriented Architectures (SOA), grid or cloud computing environments and takes care of all heterogeneity issues that arise because of differences in languages, filesystems, operating systems and hardware.

Distributed Parallel Query Engine– Clients can service queries locally or use a parallel query engine that distributes fragments of queries to simple query servers located close to the databases to be accessed. There are replaceable components for controlling how queries are split, executed and serviced. Query servers can execute filters to refine results before they are returned to the client. The query servers can be replaced with agents for accessing external data sources, such as RDBMSs or the Web.

Predictable Scalability -The address space available within a single logical view of a distributed federation of databases covers billions of Terabytes. Objectivity/DB customers have built the World’s largest databases, including a Petabyte+ system at Stanford Linear Accelerator Center. It has been used in data fusion systems that ingest and correlate over one Terabyte of data per hour with simultaneous deletion and queries.

Panasas Parallel Storage Overview

The Panasas® Parallel Storage Cluster is the premier storage system for scalable Linux clusters. Built on an object-based storage clustering architecture and an innovative parallel file system (PanFS), Panasas parallel storage delivers exceptional scaling in capacity and performance while extending appliance-like ease of management to a virtually boundless storage system. Panasas targets high-performance computing applications using Linux clusters. Government sciences, oil & gas, CAE manufacturing, life sciences and emerging commercial uses of compute clusters are the initial vertical segments.Panasas, Inc. was founded in 1999.

The main architectural featuresof Panasas ActiveStor Storage and the PanFS™ parallel file system include: ·The “single logical view” of data for simple administration and access to data. ·Easy to use, clustered distributed file system embedded on a storage appliance. ·Distributed parallelfile system and access protocol,DirectFLOW® protocol. ·Object-based storage provides a superior architecture for scalable performance.

Attributes for superior scalability:Panasas parallel storage utilizes a single global name space allowing seamless manageability at massive scale from 10 TB to multiple PB’s. Panasascustomers have successfully deployed large capacity implementations, including a 2 Petabyte shared file system at Los Alamos National Lab.

Copyright © Objectivity & Panasas 2008

2 of 7

An Objectivity and Panasas White Paper

Panasas ActiveStor Benchmark

Reliability:Panasas parallel storage is known for its superior reliability and data availability. Thepatented software RAID architecture and Tiered Parity technology ensures no single point of failure and superior data recovery on RAID protected data.

Benchmark Software

The NetElem Suite The object class model for the Network Element Management System application is shown below. Only the circled types are used in the benchmark.

Network Element Management System object class model.

The benchmark has three phases: ·Create- Creation of a tree of network elements. ·Navigate- Navigation of the tree. ·Scan- Scanning one class of objects sequentially.

Copyright © Objectivity & Panasas 2008

3 of 7

An Objectivity and Panasas White Paper

NetElem Create Phase

Panasas ActiveStor Benchmark

The application loops 32 times creating 6 containers in each loop. It creates a total of 16 GB of data comprising 86,482,055 objects of four types: ·6,144,000 Circuits ·43,395,942 Nodes ·78,113 Parts ·36,864,000 Links

The object instances are linked together using Objectivity/DB 1-to-many or many-to- many associations. The Links are a type of object, not a database relationship type.

NetElem Navigate Phase The navigation function starts from a root node and navigates to all node type objects, using Objectivity/DB persistent relationships (associations). The number of objects visited during navigation is 43,395,942, i.e. about half of the objects in the database. NetElem Scan Phase

The scan function picks the 32 Circuits containers and scans them for Circuit objects. Scanning in this case is mostly sequential read, since the container is created once and never modified. The number of Circuit objects read during the scan is 6,144,000.

Benchmark Hardware

Objectivity/DB was configured as a standalone application running in a multi-core, multi- CPU server and accessing its local disk system and the Panasas ActiveStor 5000.

The Panasas ActiveStor equipment was configured with a single AS5000 storage shelf with 1 DirectorBlade and 10 StorageBlades running ActiveScale 3.0.7. Eight client nodes were used for the benchmarking, each with a 2.8 GHz dual core Xeon and 1 GB of memory. Theclient OS was running Red Hat Enterprise Linux 4, kernel 2.6.9.

Benchmark Results (all times are in seconds)

Comparative

The first part of the benchmark compares results obtained on the local disk and the Panasas ActiveStor 5000. Note the dramatically faster navigation and scan speeds obtained with the Activestor storage configuration.

Copyright © Objectivity & Panasas 2008

4 of 7

An Objectivity and Panasas White Paper

Benchmark Step 1. Creation (Random write) 2. Navigation 3. Scan

Local (Seconds) 1648 1058 84

Panasas ActiveStor Benchmark

ActivStor 5000 (Seconds) 1241 233 10

Local Filesystem versus Panasas ActivStor 5000

1800 1600 1400 1200 1000 800 600 400 200 0 1. Creation2. Navigation (R andomw rite)

3. Scan

Local (Seconds) ActivStor (Seconds)

Scalability The second part of the benchmark measured the effect of adding processes and processors. The graph and chart below show the throughput in Megabytes per second with 1 to 8 processes running on 1 and then 2 processors.

# of Clients 1x1 2x1 4x1 8x1 8x2

WRITE BW (MB/sec) 13.2 26.3 52.1 102.2 122.7

Copyright © Objectivity & Panasas 2008

READ BW (MB/sec) 70.3 131.6 188.3 215.9 207.7

5 of 7

An Objectivity and Panasas White Paper

Panasas ActiveStor Benchmark

Bandwidth (MB/Second) Scalability

250 200 150 100 50 0 Read BW (MB/Sec) 1x1 Write BW (MB/Sec) 2x1 4x1 8x1 8x2

Write BW (MB/Sec) Read BW (MB/Sec)

The write rate increases linearly and doesn’t reach a plateau with the number of processes and threads available. The read rate scales well until the plateau is reached. Increasing the number of processors actually decreases the overall read speed, probably because of data contention, i.e. one or more processes were waiting for database locks. Summary

The benchmark convincingly demonstrated the superior performance that can be achieved with Objectivity/DB using the Panasas parallel storage system versus a local SATA disk. Writing data was up 33% faster and reading data randomly or sequentially was four to eight times faster, depending on the number of processors deployed.

Clearly, the combination of Objectivity/DB and Panasas ActiveStor 5000 offers breakthrough performance and cost advantages for applications that manipulate or query large amounts of complex data.The benchmarked configuration is particularly attractive for applications with extremely large storage requirements. The diagram below shows a highly scalable, distributed Objectivity/DB and Panasas ActiveStor configuration.

Copyright © Objectivity & Panasas 2008

6 of 7

An Objectivity and Panasas White Paper

Panasas ActiveStor Benchmark

Objectivity/DB distributed federation with local ActiveStor servers

Contact Information:

Panasas US Headquarters: Panasas, Inc, 6520 Kaiser Drive, Fremont, CA 94555 [e-mail] (888)PANASAS or [phone] Fax: [phone]

Copyright © Objectivity & Panasas 2008

. Objectivity

Objectivity US Headquarters: Objectivity, Inc. 640 W. California Ave., Suite 210 Sunnyvale, CA 94086 [e-mail] [phone] Fax: [phone]

7 of 7