Experiences with OGSA-DAI : Portlet Access and Benchmark Deepti Kodeboyina and Beth Plale Computer...
-
Upload
cecily-pitts -
Category
Documents
-
view
212 -
download
0
Transcript of Experiences with OGSA-DAI : Portlet Access and Benchmark Deepti Kodeboyina and Beth Plale Computer...
Experiences with OGSA-DAI : Portlet Access and
Benchmark
Deepti Kodeboyina and Beth PlaleComputer Science Dept.
Indiana University
Portals for Grids
Portals as user-oriented interfaces to the grid. They are used for providing user access to grid services
Potential for data access, integration, collaborative data analysis of large datasets. Data access to either commercial or scientific data would add value to the day-to-day operations of grid users.
OGSA-DAI – collaboration of grid services for providing access to heterogeneous data sources (exactly what we need)
Our goal – creation of a portlet to provide an interface for data management of disparate and heterogeneous data sources using OGSA- DAI and to understand the overheads associated with that.
Portal infrastructure
The portlet is currently deployed onto the Alliance portal that is hosted on a Linux server.
Portlet interaction
Portlet interaction return results
Provided Factory Service Handle
GDS Create service document assigned
Perform document chosen
Response as XML and html
Perform document sent to the GDS. GDS replies with response document.
Portlet interaction: asynchronous requests through gFTP
Gftp portlet
Ogsa-dai portlet
Request doc, proxy cert
gFTP gFTP
response, proxy certresponse set
On-going work
Transfer of data by means of other mechanisms (grid ftp) for providing secure access.
Transparency of the perform document location to be added. By way of using WebDAV in order to remove existing usage
restrictions.
Dynamic creation of perform documents via the portlet itself. Plan is to allow user to enter/choose required information in a web form
and dynamically generate the XML documents from that.
IU RGRbench IU RGR Benchmark – benchmark/workload for grid
information servers. realistic grid information server specific questions Synthetic workloads to test query response time under multi-client
workload Applied benchmark to analyze performance of OGSA-DAI.
quantify overhead of grid service interface to database. Benchmark queries
13 queries, iterate, sequential access to database. Data size of result set varies from few bytes to 45 KB
Benchmark scenarios synthetic workloads evaluate sensitivity of query response time to
update load. Query workloads run for 20 minutes; concurrent updates begun at
minute 6, run through minute 11. Response times averaged over interval.
Experiment details Platform Specifications
Globus toolkit 3.0 version and OGSA_DAI v3.0 deployed on linux server Database accessed: mySQL
situated on a separate dual processor Xeon server All these machines are part of the Local Area Network of the CS
department.
Experiment summarization Average query response times vary from 100 ms to 600 ms for
synchronous requests. Query response times vary with the size of the data sets also. Scenarios have varying response times extremely sensitive to the no.
of updates.
Ogsa-Dai v3.0 Query Results
Note that scale is 10^4
Synchronous response mechanism Asynchronous response mechanism
Ogsa-Dai v3.0 Scenario Results
Multiple Attribute Updates Insertion of records
Future work
On-the-fly creation of Grid Data Services in response to multiple users is being explored.
Also, when the users require data from resources which can not be created by existing factories, new ones to access these will have to be created.
More profiling being done to determine and understand the latencies
caused at a finer granularity.
Aiming at a much more broader usage of the portlet