Evaluating the Information Power Grid using the NAS Grid Benchmarks zy Rob E Van der Wijngaart* and Michael A. Frumkin NASA Advanced Supercomputing Division M/S T27A-1, NASA Ames Research Center Moffett Field, CA 94035-1000 zyx {wijngaar,frumkin}@nas .nasa.gov Abstract zyxwvuts The NAS Grid Benchmarks (NGB) are a collection of synthetic distributed applications designed to rate the per- formance and functionalit?, of computational grids. We compare seveml imp!ementdons of the NGB to deter- mine programmabilig and eficiency of NASA's Infom- tion Power zyxwvutsr Grid (IPG), whose services are mostly based on the Globus Toolkit. We report on the overheads involved in porting existing NGB reference implementations to the IPG. No changes were made to the component tasks of the NGB. guarded optimism that IPG may be beneficial for NASA's application workload, but programmability and reliability can still be improved. The rest of this paper is structured as follows. In Section 2 we briefly review the NGB, including the non-IPG ref- erence implementations. We a!so discuss some other cwId benchmarking and monitoring projects. In Section 3 we de- scribe the software and hardware inkastructure of IPG rel- evant to our experiments. In Section zyx 4 we give important details of our actual IPG NGB implementations. We dis- cuss performance and propmmability results in Section 5. Some concluding remarks about the current state of the IPG, zyxwv as well as recommendations for enhancement, are pre- sented in Section 6. 1 Introduction 2 Background As computational grids are gaining more acceptance and prominence. tools are required zyxwvutsr tn determine which rnmpn- , nents of ,gids function well and which require improve- ment. To do th2s in a systematic way, a staqdard rating mechanism must be developed, i.e. grid benchmarks. Our approach is to develop such benchmarks primarily to serve ,gd users. since increased application programmer produc- tivity and application perfomaxe are sle main goals of compiitational ,grids. Consequently, we have focused on characterizing actual distributed applications that are suit- able for execution on grids. The outcome of that work, the iirst pubiiciy avaiiabie =gid benc-bark suite, was reieaied ur?derthe nzme NAS Grid Bencbmxks (NGB), whose pre- cise specification is described in zyxwvutsr [7]. The motivation, back- ground and early experiences with NGB are reported in [?], along with a brief description of a reference implementation in Java. In this paper we discuss two implementations of NGB on NASA's production computational grid called the informa- tion Power Grid (IPG), and compare that to the earlier se- rial implementation. The performance results are reason for * Cornpurer Sciences Corpomtion Embarrassingly Distributed (ED) Figure 1. Data flow graph representing ED NAS Grid Benchmark. applications are among the simplest in terms of software in- frastructure required, and hence they should be represented in a basic +grid benchmarkins suite. The DFGs for these benchzzrks, na~ed Emba~assi~giy Dist=;buted (ED). He- lical Chain (HC), Ksualization Pipe (VP). and Mixed Bag (MB). are depicted in Figures 1 and zyxw 2. The nodes of the https://ntrs.nasa.gov/search.jsp?R=20040084431 2020-06-09T12:21:37+00:00Z