Archer:Introduction
From Grid-Appliance Wiki
Contents |
A primer on the Archer technology
Archer is based on the Grid appliance technology, which integrates three core components:
- System virtual machines (a la VMware, Xen, KVM, VirtualBox): provide a basis for packaging our software in an easy to disseminate format, and a "sandbox" for the secure execution of jobs on shared resources.
- Scalable IP virtual network (IPOP): provides connectivity among VMs deployed over the wide area Internet, dealing with constraints imposed by network address translators (NATs) in a way that is completely transparent to the end user.
- Scalable and robust job scheduler (Condor): provides the ability to submit very large numbers of independent jobs to the infrastructure, with libraries supporting job checkpointing for fault tolerance.
These technologies come together in the Grid appliance, enabling easy deployment of Wide-area Overlays of virtual Workstations (WOWs). WOWs behave functionally like a LAN cluster or NOW, but resources are virtualized and distributed across different Internet domains.
With Grid appliances, any individual researcher, student, or educator who joins Archer can:
- Quickly learn how they can submit typical computer architecture jobs to Archer nodes with self-learning modules - making the system easily accessible to undergraduate and graduate students
- Easily turn their local computer resources, from desktops to servers to clusters, into a batch-scheduled pool with minimal configuration and administration needed - all software needed to achieve this is conveniently packaged in our plug-and-play VM image.
- Easily tap into a wide-area high-throughput computing infrastructure aggregating resources from the architecture community - a seed infrastructure with hundreds of cores is being bootstrapped starting in 2008.
- Contribute with your resources to the community – transparently, securely and only when they are idle.
Archer deployment modes
Recognizing that the the infrastructure will be used in different ways by different kinds of users, Archer will support three main deployment modes: Express, Global and Local.
Archer Express
The main purpose of Archer Express is to make it very, very simple for a user with little background on VMs, Linux Condor to get started with a hands-on experiment with Archer in less than 30 minutes. It is intended for:
- Deployment on desktops and laptops
- Users trying out the system for the first time
- Students working on short-term class assignments
Archer Express includes a small-scale public resource pool, and a self-configuring VM appliance that does not require registration and authentication with Archer.
Archer Global
The main purpose of Archer Global is to serve as the production, large-scale resource pool for the computer architecture community at large. It is intended for:
- Harnessing cycles from desktop/PC labs (e.g. Linux Condor pools on Windows), clusters and servers
- Researchers seeking access to high-throughput resources from their own desktops
- Researchers seeking to easily deploy local resource pools and augment them with a global high-throughput infrastructure
- Members of the community wishing to donate cycles to Archer
Archer Global is a private pool connecting resources of registered Archer users. Unlike Archer Express, where the key goals are quick deployment and ease of use, Archer Global strives to achieve scalability with strong security guarantees. Each Archer Global VM will be authenticated: only those VMs which have host keys cryptographically signed by Archer management will be able to join the pool, and all communications among Archer VMs is encrypted end-to-end.
Archer Global provides pre-packaged software that enables you to quickly deploy a local Condor pool on top of your existing resources. When capacity in your local pool is exhausted, or if your group does not have many local computers, seamlessly use external Archer resources to run your simulations. Building on virtualization technology and the robust Condor scheduler, machines connected to Archer Global on your site will be able to "give back" to the community, without requiring extra configuration from your part, and with guarantees that your own local jobs will have pre-emptive priority over external jobs.
Archer Local
The main purpose of Archer Local is to allow users who wish to deploy an Archer pool that is fully decoupled from Archer Global and Archer Express. It is intended for:
- Harnessing cycles from desktop/PC labs (e.g. Linux Condor pools on Windows), clusters and servers
- Enterprises seeking to deploy and manage Archer software within their private networks
With Archer Local, you can be in control of the entire configuration and management of your local cluster, independently from other Archer resources. This comes with advantages and drawbacks:
- In Archer Local you can define and implement your own authentication and scheduling policies and be assured your resources will only run jobs from your users
- However, you will not be able to "spill out" jobs to external resources as in Archer Global
- In Archer Local you will be able to reuse the Archer software stack, which is open-source
- However, you will be responsible for management, whereas Archer Global is managed for you
Deployment status
Archer Express: Summer 2008: the public Archer Express infrastructure has been deployed, as well as the Wiki with documentation, videos and how-tos for various popular architecture simulators including Simics, SESC and SimpleScalar.
Archer Global: Fall 2008: the first deployment of Archer Global on resources at U. Florida. Additional 336 cores to come online (at FSU, UT Austin and Northeastern).
Archer Local: documentation on how to set up local Grid appliance pools is available in tutorials linked from the main Grid appliance site. A management VM appliance that will facilitate configuration and management of Archer local pools is slated for deployment in the Spring of 2009.
