G. Heiser, F. Lam, S. Russell, Resource management in the Mungi single-address-space operating system, in: Proceedings of the Australasian Computer Science Conference, Perth, Australia, February 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Sharing Model . Community Coalition Bartering Model . Monopoly and Oligopoly Several research systems (see Table 1) have attempted to apply the concept of computational economy models to database management, CPU cycles, storage, and distributed computing. They include Mariposa [8] Mungi [17], Popcorn [21] Java Market [18] Enhanced MOSIX [19] JaWS [30] Xenoservers [31] D Agents [32] Rexec Anemone [22] Spawn [20] Mojo Nation [24] and Nimrod G [1] 2] These systems have been targeted to manage single or multiple resources and they can be categorized as follows: Single ....

....has been discussed in the next section. SYSTEM NAME ECONOMY MODEL PLATFORM REMARKS Mariposa [8] UC Berkeley) Bidding (Tendering ContractNet) Pricing based on load and historical info. Distributed database. It supports budget based query processing and storage management. Mungi [17] (University of New South Wales) Commodity market (renting storage space that increases as available storage runs low, forcing users to release unneeded storage. Storage servers. It supports storage objects based on bank accounts from which rent is collected for the storage occupied ....

G. Heiser, F. Lam, and S. Russell, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, February 4-6,

....including microeconomic and macroeconomic principles have been proposed in the literature. Some of the commonly used 3 economic models for selling goods and services can be employed as service price negotiation protocols in Grid computing, include: Commodity Market Model (case systems: Mungi [20], Enhanced MOSIX [21] and Nimrod G [1] 4] Posted Price Models (case system: Nimrod G) Bargaining Model (case systems: Mariposa [10] and Nimrod G, Tendering Contract Net Model (case system: Mariposa [10] Auction Model (case systems: Spawn [22] and Popcorn [23] Bid based ....

G. Heiser, F. Lam, and S. Russell, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, Perth Australia, Feb. 1998, Springer-Verlag, Singapore, 1998.

....2 selling goods and services can be employed as negotiation protocols in Grid computing. Some of these market or social driven economic models are shown in Table 1 along with the identity of the distributed system that adopted the approach [8] Economic Model Adopted by Commodity Market Mungi [16], MOSIX [17] Nimrod G [1] 4] Posted Price Nimrod G Bargaining Mariposa [12] Nimrod G Tendering or Contract Net Model Mariposa [12] Auction Model Spawn [18] Popcorn [19] Bid based Proportional Resource Sharing Rexec Anemone [20] Community, Coalition, and Bartering Condor, SETI Home ....

G. Heiser, F. Lam, and S. Russell, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, Perth Australia, Feb. 1998, Springer-Verlag, Singapore, 1998.

....including microeconomic and macroeconomic principles have been proposed in the literature. Some of the commonly used economic models for selling goods and services can be employed as service price negotiation protocols in Grid 3 computing, include: Commodity Market Model (case systems: Mungi [18], Enhanced MOSIX [19] and Nimrod G [1] 2] Posted Price Models (case system: Nimrod G) Bargaining Model (case systems: Mariposa [8] and Nimrod G, Tendering Contract Net Model (case system: Mariposa [8] Auction Model (case systems: Spawn [20] and Popcorn [21] Bid based ....

G. Heiser, F. Lam, and S. Russell, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, Perth Australia, Feb. 1998, Springer-Verlag, Singapore, 1998.

....overviews over past research in this area. Most of the systems described are auction based and intended for more coarse grained resource allocations. In [17] a proportional share scheduling algorithm is described in which tasks are charged for their resource usage and are refunded periodically. In [6] a system where objects are charged for backing store usage is presented. Charges are based on an analytical cost function of the storage utilisation. It is interesting to note that none of these systems base their charges on shadow prices, which have been demonstrated to be a useful mechanism to ....

G. Heiser, F. Lam, and S. Russell. Resource Management in the Mungi SingleAddress -Space Operating System. In Proc. of the 21st Australasian Computer Science Conference, Perth, Australia, February 1998.

....This model works when all participants in the Grid are both service providers and consumers. Several research systems (see Table 1) have attempted to apply the concept of computational economy for information management, CPU cycles, Storage, and Network access. They include Mariposa [19] Mungi [30], Popcorn [33] Java Market [31] Enhanced MOSIX [32] JaWS [17] Xenoservers [34] D Agents [35] Rexec Anemone [29] Spawn [36] Mojo Nation [25] These systems can manage either single or multiple resources and they are categorized as follows: Single Domain Systems: Enhanced MOSIX and ....

Heiser G, Lam F, and Russell SM, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, Perth Australia, Feb. 1998, Springer-Verlag, Singapore, 1998.

....This model works when all participants in the Grid are both service providers and consumers . Several research systems (see Table 1) have attempted to apply the concept of computational economy for information management, CPU cycles, Storage, and Network access. They include Mariposa [19] Mungi [30], Popcorn [33] Java Market [31] Enhanced MOSIX [32] JaWS [17] Xenoservers [34] D Agents [35] Rexec Anemone [29] Mojo Nation [25] These systems can manage either single or multiple resources and they are categorized as follows: Single Domain Systems: Enhanced MOSIX and Rexec Anemone. ....

Heiser G, Lam F, and Russell SM, Resource Management in the Mungi Single-Address-Space Operating System, Proceedings of Australasian Computer Science Conference, Perth Australia, Feb. 1998, Springer-Verlag, Singapore, 1998.

....is therefore an independent component in addition to the memory management and does not cooperate with it. Furthermore, segmentation makes the migration ability more difficult and is limited to very large objects. Many other projects uses a similar approach, as for example the Mungi project [10, 11], and therefore share the same disadvantages. 5 First performance results We tested the prototype implementation of the MoDis VAS with a parallelized quicksort algorithm. As a hardware configuration we used a cluster of ten SUN UltraSparc Creator One workstations connected via a 100 MBit s ....

Gernot Heiser, Fondy Lam, and Stephen Russell. Resource management in the mungi single-address-space operating system. In Proc. 21st Australasian Comp. Sci. Conf, page wird erscheinen, Februar 1998.

....support resource transfer between processes through thread tunnelling. This attempts to solve the problem of QoS crosstalk introduced by use of shared servers on the data path, but with increased complexity when compared with the Nemesis approach of performing work in the client processes. Mungi [13] makes use of an economy based model for allocation of backing storage, to support garbage collection of unwanted memory segments. In [3] Resource Containers are proposed to allow servers to perform finegrained resource management between their clients, using hierarchical resource allocations to ....

G. Heiser, F. Lam, and S. Russel. Resource Management in the Mungi Single-Address-Space Operating System. In Proceedings of the 21st Australasian Computer Science Conference, Feb. 1998.

....two different versions of sockets may exist: a standard one and one which transmits and receives compressed data. 6 A critique of economics based approach to resource management A number of research projects advocate the use of economical models to force applications to compete for resources [3,5,10]. This approach may seem ideal for Internet systems. In fact, there is an unquestionable merit to bidding for resources prioritizing applications according to the amount of funding they have, which should in turn reflect the relative importance of different tasks. Another advantage is the ....

....to compare the value or importance of two resources through comparing their price . However, there are several problems with this approach. First, the actual pricing mechanism implemented in existing and proposed systems seems to invariably be either cheap to implement but rather arbitrary [5] or more realistic auction based but with high overheads. For example, Edell et al. 3] propose a system that enables billing users for their TCP traffic but it increases the average connection setup time by 66 . A second issue is who actually funds applications in economy based resource ....

[Article contains additional citation context not shown here]

Heiser, G, Lam, F, and Russel S. Resource Management in the Mungi Single-Address-Space Operating System. Proc. 21st Australasian Computer Science Conference, Perth, Australia, February 1998.

....funding they have, which should in turn reflect the relative importance of different tasks. However, there are several problems with this approach. First, the actual pricing mechanism implemented in existing and proposed systems seems to invariably be either cheap to implement but rather arbitrary [3] or more realistic auction based but with high overheads. For example, Edell et al. 1] propose a system that enables billing users for their TCP traffic but it increases the average connection setup time by 66 . A second issue is who actually funds applications in economy based resource ....

....it increases the average connection setup time by 66 . A second issue is who actually funds applications in economy based resource management environments. If it is the system, then the accumulation of money over periods of inactivity must be dealt with. Some systems propose a form of taxation [3], but the formulas used are again rather arbitrary. If the user funds the applications based on real money the pricing and its evolution must be easy to understand. Most current services appear to quickly converge on flat rates. However, introducing flat rates actually defeats the purpose of ....

Heiser, G, Lam, F, and Russel S. Resource Management in the Mungi Single-Address-Space Operating System. Proc. 21st Australasian Computer Science Conference, Perth, Australia, February 1998.

....users and their applications to limit resource consumption. Each time an application wants to obtain rights to use a particular resouce, it has to pass its own resource group capability as a hidden argument on system calls. Another example of a single address space operating system is Mungi [21]. An interesting feature of that system is that it uses an economics based model to manage backing store management. Applications obtain bank accounts from which rent is collected for the storage occupied by objects. Rent automatically increases as available storage runs low, forcing users to ....

Heiser, G, Lam, F and Russel, S. Resource Management in the Mungi Single-Address-Space Operating System. In Proceedings of the 21 st Australasian Computer Science Conference, Perth, Australia, February, 1998.

....a quota system which limits the amount of space a user can occupy. Alternatively, economic models can be used, which associate a price with a resource, and users buy or rent space, involving some form of payment [Anderson et al. 1986, Mullender and Tanenbaum, 1986, Drexler and Miller, 1988, Heiser et al. 1998b] On the other hand, processor time is allocated according to some priority scheme. Priorities can be hard, meaning that a process will only execute if no higher priority process is runnable, or soft, meaning that a process priority in uences the frequency or duration for which its process is ....

....The design of a generic resource management framework is presented in Section 4, followed by details on the mechanism for prioritising physical memory allocation in Section 5. Experimental results for a prototype model are presented in Section 6. 2 Mungi 2. 1 Fundamental Abstractions Mungi [Heiser et al. 1998a] is a single address space operating system (SASOS) and as such executes all processes on all nodes in a single, large virtual address space. This address space contains all persistent and transient data, simplifying data sharing and persistent storage. All operations on this address space are ....

[Article contains additional citation context not shown here]

Heiser, G., Lam, F., and Russell, S. (1998b). Resource management in the Mungi single-address-space operating system. In Proceedings of the 21st Australasian Computer Science Conference (ACSC), pages 417-428, Perth, Australia. Springer-Verlag. Also available as UNSW-CSE-TR-9705 from http://www.cse. unsw.edu.au/school/research/tr.html.

....a quota system which limits the amount of space a user can occupy. Alternatively, economic models can be used, which associate a price with a resource, and users buy or rent space, involving some form of payment [Anderson et al. 1986, Mullender and Tanenbaum, 1986, Drexler and Miller, 1988, Heiser et al. 1998b. On the other hand, processor time is allocated according to some priority scheme. Priorities can be hard, meaning that a process will only execute if no higher priority process is runnable, or soft, meaning that a process priority influences the frequency or duration for which its process is ....

....The design of a generic resource management framework is presented in Section 4, followed by details on the mechanism for prioritising physical memory allocation in Section 5. Experimental results for a prototype model are presented in Section 6. 2 Mungi 2. 1 Fundamental Abstractions Mungi [Heiser et al. 1998a] is a single address space operating system (SASOS) and as such executes all processes on all nodes in a single, large virtual address space. This address space contains all persistent and transient data, simplifying data sharing and persistent storage. All operations on this address space are ....

[Article contains additional citation context not shown here]

Heiser, G., Lain, F., and Russell, S. (1998b). Resource management in the Mungi single-address-space operating system. In Proceedings of the 21st Australasian Computer Science Conference (A CSC), pages 417-428, Perth, Australia. Springer-Verlag. Also available as UNSW-CSE-TR-9705 from http://www. cse. unsw. edu. au/school/research/tr. html.

....in more detail in the Work in Progress Section below. Storage management While the kill list helps to reduce the amount of garbage objects, this is not enough to prevent all secondary storage eventually filling up with unused objects. Automatic garbage collection does not provide a solution [24]. As in a traditional file system, persistent objects are normally entered into a directory, which associates human readable names with 64 bit object addresses. As long as the directory continues to contain a reference to an object, it cannot be automatically removed as garbage. The system, ....

....to clean up. Note that in such a situation users who have let their account balance drop low will be the first to run out of funds, while those using storage economically will be less affected. Such a graceful degradation of service cannot be achieved with a quota system. Details can be found in [24]. It should be pointed out that Mungi s support for bank accounts is limited to requiring the presentation of a valid bank account capability at object creation time. The details of the accounting model are implemented at user level. Hence it is easy to change the accounting policy to something ....

G. Heiser, F. Lam, and S. Russell. Resource management in the Mungi single-address-space operating system. In Proceedings of the 21st Australasian Computer Science Conference, pages 417--428, Perth, Australia, February 1998. Springer-Verlag. Also available as UNSW-CSE-TR-9705 from http://www.cse.unsw.edu.au/school/research/tr.html.

No context found.

G. Heiser, F. Lam, S. Russell, Resource management in the Mungi single-address-space operating system, in: Proceedings of the Australasian Computer Science Conference, Perth, Australia, February 1998.

No context found.

Heiser G, Lam F, Russell S. Resource management in the Mungi single-address-space operating system. Proceedings of Australasian Computer Science Conference, Perth, Australia, 4--6 February 1998. Springer: Singapore, 1998.

No context found.

G. Heiser, F. Lam, and S. Russell, "Resource management in the mungi single-address-space operating system," presented at the Australasian Computer Science Conf., Perth, Australia, Feb. 4--6, 1998.

No context found.

Heiser, G., Lam, F., and Russell, S. Resource management in the mungi single-address-space operating system. In Proceedings of the 21st Australian Computer Science Conference (February 1998).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC