Home

GSSIM

Grid simulation tools provide frameworks for simulating application  scheduling in various Grid infrastructures. However, while experimenting  with many existing tools, we have encountered two main shortcomings:

• lack of tools for generating workloads, events and resources;
• it is difficult and time consuming to model different Grid levels i.e. resource brokers, and local level scheduling systems

The Grid Scheduling Simulator (GSSIM) is a framework that addresses these  shortcomings and provides an easy-to-use Grid scheduling framework for enabling  simulations of a wide range of scheduling algorithms in multi-level, heterogeneous  Grid infrastructures. GSSIM is built on top of GridSim.

GSSIM has been designed as a simulation framework which enables easy-to-use  experimental studies of various scheduling algorithms. Its goal is also to allow  researchers to move the implementations of scheduling algorithms between simulation  environments and real systems. GSSIM supports multilevel scheduling architectures  with plugged-in algorithms both for Grid and local schedulers.  It also enables both reading existing real workloads and generating synthetic  Grid workloads based on given probabilistic distributions and constraints.  These workloads are compliant with known workload formats such as  Standard Workload Format (SWF) and Grid Workload Format (GWF).  The framework also supports generation of resource failures which may be useful  in modeling realistic behavior of Grid environments. The GSSIM framework is based on GridSim and SimJava2 packages.  It provides a layer added on top of the GridSim adding capabilities  to enable easy and flexible modeling of Grid scheduling components.  GSSIM also provides an advanced generator module using real and synthetic workloads.  The overall architecture of GSSIM is illustrated in the following figure:

GSSIM distinguishes between two types of scheduling components:  Grid brokers and resource providers.  As shown in the figure above, multiple scheduling strategies may  be plug into both levels. Input data can be read from real sources or generated  using the generator module.

Motivations and Goals

Availability of measurable and comparable results is very important in evaluation  of algorithms and strategies for Grid scheduling.  For this purpose there are activities aiming at providing access to reusable  workloads such as the Grid Workloads Archive and the  Parallel Workloads Archive Nevertheless, these activities focus mostly on repositories of real workloads.

To make another step forward in sharing and comparing the results in Grid  scheduling research we believe that we could also share implementations  of algorithms and synthetic workloads. Therefore, we created this  Grid Scheduling Simulations portal.

In many research papers new proposed Grid scheduling algorithms are  compared against other well-known approaches or algorithms published by other authors. These comparisons would be more reliable and easy if existing implementations  of algorithms could be reused. This is possible in GSSIM where they are inserted  as plugins to the simulation infrastructure. The main goal is to have both Grid  and local scheduler plugins implementing various scheduling algorithms available  at this portal.

Workloads coming from real Grid environments are still hardly available.  Even if they are available they have certain drawbacks and limitations.  In all these situations use of synthetic workloads may be indispensable.  Especially as use of synthetic workloads has been positively investigated in several papers. For all these reasons, we believe that publishing the synthetic workloads may be useful. Among other advantages, it will allow researchers to make repeatable experiments and refer  to specific workloads used in their research.  They can download existing workloads from GSSIM portal or generate a new workload based on specified parameters using an online workload generator available also at the portal.  In addition to this, users of the GSSIM simulator may take advantage of implementations of  Grid and local level scheduling algorithms.

Repositories

The GSSIM portal contains repositories of all elements needed to perform scheduling experiments with GSSIM. Categories of these elements are as follows:

• Workloads
• Description of resources
• Plugins