The Center for Information Technology Integration (CITI) is studying
the performance of the proposed
M-Pathways infrastructure,
which is based on a distributed processing architecture provided by
PeopleSoft. The CITI performance
study consists of two complementary efforts. The first involves
building a testbed at CITI to measure the performance of specific
infrastructure components. The second effort involves modeling the
infrastructure as a whole, and seeks to provide overall performance
indicators.
Testbed Effort
The CITI M-Pathways Performance Measurement testbed is modeled after the
current M-Pathways deployment architecture and consists of a three-tier
hardware architecture:
-
Server: IBM RS/6000, Model F30, AIX 4.1.4, 256 MB RAM, 4 GB disk
-
Middle: Pentium, dual-processor, 200 Mhz, Windows NT, 128 MB RAM, 4 GB disk
-
Client: Pentium, 200 Mhz, Windows 95/NT, 32 MB RAM, 2 GB disk
The IBM RS/6000 runs Oracle7 version 7.3.2.3 server code and is the
same Oracle code that is invoked by the PeopleSoft 6 software.
The dual-processor Pentium runs the Citrix WinFrame 1.6 server
software; the single-processor Pentium runs the WinFrame client.
Together, the WinFrame product allows an application's user interface
to execute on a client PC while application logic executes on the
WinFrame server.
Stress, a program to stress the Oracle and Citrix products in a
controllable way, was developed in Visual Basic and has been used to
evaluate both two- and three-tier architectures. In the two-tier
configuration, Stress is run on the dual-processor Pentium and accesses
the Server machine directly; in the three-tier configuration, Stress
is invoked from the WinFrame client machine, runs on the WinFrame
server, and passes screen updates back to the WinFrame client under
control of the Citrix product.
Currently, we're writing a report summarizing our findings.
Modeling Effort
A closed queueing model (QNM) with a fixed number of customers was built
to analyze the two-tier configuration. Stress was used to obtain values
for model parameters reflective of our testbed configuration, and the
model is now being used to generate performance predictions.
Currently, we're writing a report summarizing our results.
Future Work
The next step involves adding a third tier to the performance model.
