The Grid presents significant new challenges for organizational security. Typical Grid applications require complex and dynamic patterns of trust to be established and to be implemented by computers making decisions about the identity and rights of individual users. In addition, much of the Grid middleware in common use has been designed for environments in which there is a high level of trust between systems and there is an implicit assumption that the network infrastructure is very open. Within such an environment, the major concern is to ensure that only legitimate users can access the set of resources to which they are entitled. There is often the tacit assumption that would-be illegitimate users are excluded by some external perimeter surrounding the enclosed trusted environment. Therefore, much of the concern with Grid security has been concentrated on authentication of users and authorization of resources within such a trusted environment rather than focusing on attacks that may come from outside.
Most university networks are moving away from a simple security model in which everything inside a perimeter is trusted and everything outside is not. This move toward a more complex security structure is due to two pressures: (1) the need for inter-institutional collaboration with people and systems outside such a perimeter and (2) the increased threat from rapidly spreading worms and viruses that can appear both outside and inside the perimeter. Production networks in education now typically implement firewalls and routers that control the flow of traffic both between and within organizations.
The problem is that these security measures were designed to manage traditional patterns of computing in which client and server functions are performed by different systems with well known protocols and where there are simple trust relationships reflecting the organizational hierarchies. By contrast, Grid applications often require transmission of very large amounts of data at high speeds in non-standard use patterns. This can cause problems for these routers and firewalls. For example, firewalls need to be able to distinguish between “good” traffic that should be permitted to pass and “hostile” traffic that should be blocked. Traditionally, this has been done using port numbers with a typical firewall allowing connections on ports 80 and 443 to the Web server and port 25 to the mail server. For simple protocols using a single TCP connection, this level of description by port and destination is adequate. Grid applications, however, tend to involve multiple connections between groups of machines and these protocols are much more complex to describe.
Of course, for Grids deployed entirely within an enterprise, security is perhaps not such an issue. However, even here there is the need for reliable accounting solutions, as Grids typically cross multiple administrative domains. More fundamental security issues arise when Grid applications cross institutional boundaries. Although the move toward Web services is to be welcomed as a step toward building robust, industrial-strength Grid middleware, the situation surrounding security standards is, unfortunately, confused.
It was for this reason that a meeting on “Security for e-Science: Approaches and Interoperability” was held April 11 in London by the U.K. e-Science Core Program. Speakers representing the various different approaches made presentations on the possible key technologies and the meeting was very useful in presenting a “map” of the present state of Grid security. However, as can be seen from the following summary of the talks at this meeting, there are many different strands that must ultimately be reconciled to provide an acceptable set of Grid security solutions.
The meeting opened with an overview of security technologies from Peter Henderson of the United Kingdom's Open Middleware Infrastructure Institute. Marty Humphrey from the University of Virginia then gave a presentation on the present sorry state of Web service security standards. He began by reminding the audience of the beautiful Web services security roadmap proposed by IBM and Microsoft in 2002.
From this roadmap, the WS-Security proposal has progressed from a proposed specification in April 2002 to an OASIS standard by April 2003. Unfortunately, the status of other specifications on the roadmap — such as WS-Policy, WS-SecureConversation, WS-Trust and WS-Federation — is rather unclear. The statuses of WS-Privacy and WS-Authorization are even less clear. Humphrey then described the WS-I Security Profile and the SAML and XACML OASIS standards and noted a potentially confusing overlap between XACML and SAML. He concluded with a discussion of WS-Delegation, a GGF standards effort led by Olle Mulmo. David Chadwick then gave an overview of GGF activities on authentication and authorization using PERMIS. He stressed that it is the use of resources by an application — rather than a user — that is the primary motivation for introducing proxy certificates.
From these rather abstract discussions, the meeting moved on to concrete implementations, with talks from Frank Siebenlist on the Globus GT4 security architecture, Olle Mulmo on the approach to security of the European EGEE project and Li Zha on security in the CNGrid VEGA project. Siebenlist began by stressing the importance of a security policy for a virtual organization (VO) and the role of security services in facilitating enforcement of this policy. After reviewing the OGSA security services, he outlined GT4's support for different mechanisms for attribute assertions — VOMS, PERMIS, X.509, Shibboleth and SAML. He noted that moving to an SAML/XACML approach provided GT4 with a more flexible way to support a variety of security solutions than just X.509 certificates. An open source XACML runtime is being shipped with GT4. He concluded that the capability to form a dynamic, “five-minute VO” is still an as yet unrealized goal for the Grid community.
Mulmo from KTH in Stockholm then described the EU-funded EGEE project that connects more than 100 sites across Europe and constitutes probably the largest production Grid in the world with more than 10,000 CPUs and over 5 petabytes of storage. The EGEE infrastructure uses X.509 and proxy certificates with the VOMS package for authorization and VO management. He noted that for the third “A” of AAA, “accounting,” no solution was currently deployed. Similarly, there was as yet no overall solution for “audit,” the so-called fourth “A” of AAA.
Zha from the ICT Institute of the Chinese Academy of Sciences described security in the VEGA project. After an overview of the VEGA GOS architecture, he described the Agora and Grip services that, along with WS-Security, form the basis for implementing the VEGA security architecture. The Agora service provides VO authorization and access control using a SAML based authorization token. The Grip service is a Grid process that provides a runtime construct to deliver secure access to a service.
So far, the meeting had concentrated on the security concerns of a VO as typified by Grid application projects. However, members of any Grid VO are also members of institutions, each of which have their own security infrastructure. The remaining talks were concerned with Internet2's Shibboleth architecture. Shibboleth mandates neither a specific authentication scheme (this is taken to be the responsibility of the home institution) nor a specific authorization scheme (this is the responsibility of the resource owner). Instead, Shibboleth is an open standards-based protocol for securely transferring attributes between the home site and the resource site. The message flows are defined in SAML and the institution member attributes (in U.S. academia) are typically taken from the eduPerson and eduOrg schemas.
There is growing international acceptance for Shibboleth with its deployment by some 30 major U.S. universities; serious trials in the United Kingdom, Switzerland and other countries in Europe; and a significant deployment project in Australia. Alan Robiette from JISC reported on an interesting U.K. project attempting to add more intelligence to the resource manager by integrating PERMIS as a decision engine within the Shibboleth framework. Martin Sutter from the Swiss academic research network organization SWITCH reported on the “SWITCHaai” project. This has successfully deployed a Shibboleth-based Authentication and Authorization infrastructure (AAI) across the seven major Swiss universities. The final talk of the day was by Von Welch from NCSA who explained the goals of the NSF-funded “GridShib” project. This project aims to incorporate Shibboleth-issued attributes for authorization into Grids built on the Globus GT4 toolkit. The project is exploring possible methods of managing attributes for members of Grid VOs.
Although the picture for Grid security is a very complex one, the meeting was useful in exposing some of the real practical problems facing Grid VOs. I conclude this tale of unfinished business by describing the U.K. “GOLD” e-Science research project that is examining even more security issues for VOs. The project is focusing on the problems of coordination, information management, security and trust in a VO. The project is grounding its work with a proof of concept example VO taken from the chemical industry. The GOLD work on coordination and trust has focused on distributed workflow enactment. It has taken a more formal approach to these issues than most Grid projects and is defining semantics using the pi calculus and using the SPIN model checker for compositions. In addition, because its VOs involve commercial organizations, legal issues such as rights and obligations need to established through some form of electronic contracts. The project is using the Promela language to specify such contracts and implementing a “two-phase commit” type protocol to enforce non-repudiation.
As can be seen from the wide range of issues touched on in this article, there is still much to be done on Grid security!
A UKERNA Guidance Note on “Deploying Grids” by Andrew Cormack, UKERNA's chief security adviser, is available on the UKERNA Web site at www.ukerna.ac.uk. The presentations at the U.K. e-Science Security Meeting can be found on the NeSC Web site at www.nesc.ac.uk/events/townmeeting0405.
© Tony Hey May 2005