D. McNamee and K. Armstrong. Extending the mach external pager interface to accomodate user-level page replacement policies. Technical Report 90-09-05, University of Washington Dept. of Comp. Sc. and Engg., 1990.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... We consider this overly restrictive because this policy is always a compromise between performance and generality; related work has shown that applications are often ill served by the default operating system policy [1, 50] and can benefit significantly from managing their own memory resources [15, 19, 20, 27, 29, 39]. Also, while the allocation policy can be configured in some of the approaches, other policies, e.g. for placement or replacement, cannot be influenced at all. Furthermore, the effectivity of this approach depends strongly on the policy that is being used. FCFS, which is implemented in L4 and ....

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. Technical Report TR-90-09-05, Department of Computer Science and Engineering, University of Washington, 1990.

....and especially the management of the execution structures. Protected ports have a major drawback in a distributed environment: applications cannot share ports. Port group would have been convenient for managing distributed entities or providing fault tolerance facilities. As proposed in [McNamee 90] it would be interesting to provide the ability to manage page replacement in physical memory at the level of a memory managers, for the improvement of cluster paging. 4.3. Perspectives The Guide system is currently used for the development of new experimental applications, and as a ....

D. McNamee and K. Armstrong, Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies, Proc of the Mach Usenix Workshop, Burlington, VE, Oct 1990, pp 31-43

....requests for a particular resource. 2.1 Approaches to customized resource management A number of systems have attempted to remedy this problem by exposing greater resource control to applications. Scheduler activations [7] application specific handlers [143] user level virtual memory managers [58, 93], and operating systems such as SPIN [16] Exokernel [70] and Nemesis [87] are all attempts to augment limited operating system interfaces by giving applications the ability to specialize the policy decisions made by the kernel. Application specific handlers allow applications to push fast ....

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the USENIX Mach Symposium, 1990.

....Implementation of synchronisation tools such as semaphore objects shared between task located on the same or different nodes is not an easy work. The designer should either provide a semaphore server or ensure that all threads sharing a semaphore object know each other s ports. As proposed in [McNamee 90] it would be interesting to provide the ability to manage page replacement in physical memory at the level of a cluster manager, since a cluster manager does have some knowledge about cluster usage that the kernel cannot manage. For example in the current implementation, cluster managers collect ....

D. McNamee and K. Armstrong, Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies, Proc of the Mach Usenix Workshop, Burlington, VE, Oct 1990, pp 31-43

....we should know the capacity of a block and the internal memory beforehand, 2. we must be able to align elements into blocks, and 3. we must have a full control over the replacement of the blocks in internal memory. There are operating systems that provide support for these facilities (see, e.g. [17, 21, 23]) The tree structure of Arge and the heap structure of Fadel et al. do not give any guarantees for the performance of individual operations. In fact, one Insert or DeleteMin can be extremely expensive, the cost of handling the whole sequence being an upper bound. Therefore, it is risky to use ....

D. McNamee and K. Amstrong. Extending the Mach external pager interface to accommodate user-level block replacement policies. Technical Report 90-09-05, Department of Computer Science and Engineering, University of Washington, Seattle, 1990.

....obvious that sequence detection can be used for prefetching as well. How to balance the prefetching and caching, though, is a complicated issue that needs further study [7] Recent research projects on application controlled kernels show the potentials of application specific replacement policies [34, 17, 25, 23]. However, most of those studies focus on letting applications inform kernel of their access patterns, instead of having operating systems detect them at run time (as in the SEQ algorithm) Our study shows that in fact, runtime automatic detection by kernel is feasible and can perform quite well. ....

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of USENIX Mach Symposiumi '91, pages 17--29, 1990.

....6 Related Work Numerous efforts have been made to provide application specific VM. Mach [25] user level pagers allow applications to control how data is transferred between physical memory and backing store. User level pagers were later incoporated by Chorus [1] and Spring [13] Premo pagers [20] and extensible object oriented virtual memory [14] extended Mach pagers by allowing pagers to implement replacement policies. HiPEC [15] allows applications to control replacement policies by downloading policies written in a restricted language to the kernel. Sechrest [24] and V page cache ....

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accomodate user-level page replacement policies. In Mach Workshop. USENIX, October 1990.

....may not be either optimal for the reference strings that are encountered or identical to the stack buffer replacement policy. This can be overcome by providing operating system kernel mechanisms that the user level buffer manager can use to realize applicationspecific memory replacement policies [18, 19] that are more appropriate. If memory mapping facilities can be improved in this way, then it will be possible to implement an arbitrary symmetric maximal pool system using the memory mapped approach. Otherwise, only those symmetric maximal pool systems whose buffer replacement policy is identical ....

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to accommodate User-level Page Replacement Policies. Proc. of the USENIX Association Mach Workshop, Burlington, Vermont, October 1990, pp. 17-29.

....to improve their performance has been suggested before. For instance, the Mach OS supports external pagers to allow applications to control the backing storage of their memory objects [18] Extensions to the external pager interface have been used to implement user level page replacement polices [14], and to support discardable pages (i.e. dirty pages that do not need to be written to backing store) 20] In contrast, our approach shows that specialized applications can and should exploit extra control for the benefit of other applications executing concurrently. This is especially true for ....

....this functionality to the applications. For instance, the Mach operating system supports external pagers to allow applications to control the backing storage of their memory objects [18] Extensions to the external pager interface have been used to implement userlevel page replacement policies [14] and to support discardable pages (i.e. dirty pages that do not have to be written to backing store) 20] More aggressive application control of physical memory was implemented in the V kernel by Harty and Cheriton [10] In their scheme, the application was given complete control over a cache ....

D. McNamee and K. Armstrong. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proc. of the USENIX Assoc. Mach Workshop, pages 17--29, 1990.

....reclaims unreferenced pages found by the hand. Unlike Mach s algorithm, the BSD algorithm suffers from scalability problems, because the clock hand must sweep across the machine s entire physical memory. It is also difficult to tune properly the speed 10 of the clock hand. Some researchers [McNamee Armstrong 90] have taken the approach of providing control over page replacement to the application. This is a promising approach for applications like Lisp, ML, and databases, but it leaves open the question of a default mechanism. Mach s FIFO with Second Chance algorithm can easily accommodate simple ....

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of the USENIX Mach Workshop, pages 17--29, October 1990.

....for very limited memory (M=32) remote and local are essentially at parity. 6 Related work Researchers in operating systems have recently explored extensions to standard systems to support more application control over virtual memory. The case for these extensions has been made repeatedly (cf. [2, 10, 14, 18, 21, 27]) Some research prototypes have added more application control [11, 14, 20, 21, 27] but these features have unfortunately not found their way into commercial operating systems. Appel and Li have demonstrated by operating system modification that application control over write back policies can ....

....work Researchers in operating systems have recently explored extensions to standard systems to support more application control over virtual memory. The case for these extensions has been made repeatedly (cf. 2, 10, 14, 18, 21, 27] Some research prototypes have added more application control [11, 14, 20, 21, 27] but these features have unfortunately not found their way into commercial operating systems. Appel and Li have demonstrated by operating system modification that application control over write back policies can improve performance by discarding dirty data that really are garbage or that can be ....

D. McNamee and K. Armstrong, "Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies," Proceedings of the USENIX Association Mach Workshop, 1990, pp. 17-29.

....For example, Windows NT[11] adopts FIFO replacement. Mach[1] introduced the idea of an external pager[45] software outside the operating system kernel proper, yet capable of implementing paging mechanisms. Others have extended the external pager interface to expose the replacement policy as well[29, 39]. V [18] implements a segment based scheme in which user level processes control segments. Recently even more exotic interfaces have been proposed, including one built around a so called metaobject protocol[24] and another that involves applications downloading replacement policies coded in a ....

Dylan McNamee and Katherine Armstrong. Extending the mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the USENIX Association Mach Workshop, Burlington, Vermont, October 1990. 110

....16 application kernels simultaneously. Finally, the Caching Kernel has an alarmist view of downloading code into the kernel, limiting the flexibility and efficiency of their approach. User level pagers allow a rudimentary control over the VM system by allowing decisions about which pages to swap [16]. However, their interface and power is very limited: they do not allow control over pagetable structure, allocation of specific physical pages, or even access to many page attributes (e.g. pagesize or uncached) 7] The Bridge project attempts to move the virtual memory decisions into the ....

Dylan McNamee and Katherine Armstrong. Extending the mach external pager interface to accommodate user-level page replacement policies. In Mach Workshop Conference Proceedings, pages 17--30, Burlington, VT, October 4-5 1990. USENIX.

....or remote paging whether they benefit or not. Furthermore, in the case of a global compressed caching scheme, the applications that do not benefit degrade the performance of those that do by causing the cache size to be reduced. Other developments in user level virtual memory management [18, 13, 8] and quality of service (QoS) 14, 3, 19] have provided the impetus for a more specialised and deterministic environment. The aim of this work is to take advantage of the developments in QoS and user level memory management, combining these with solutions for the disk latency problem to provide a ....

McNamee, D., and Armstrong, K. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proceedings of the Worshop on Mach (Burlington, VM, USA, October 1990), USENIX Association, pp. 17--30. 15

....in an attempt to speed up page fault handling ( 18, 9] While these solutions o er general bene ts, neither is application centric. All applications must pay the performance cost of compression or remote paging whether they bene t or not. Other developments in user level virtual memory management [21, 16, 8] and quality of service (QoS) 17, 3, 24] have provided the impetus for a more specialised and deterministic environment. My aim is to take advantage of the developments in QoS and user level memory management, combining these with solutions for the disk latency problem to provide a exible, ....

McNamee, D., and Armstrong, K. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proceedings of the Workshop on Mach (Burlington, VM, USA, October 1990), USENIX Association, pp. 17{ 30.

....neither is application centric. All applications must pay the performance cost of compression or remote paging whether they benefit or not. Furthermore, the applications that do not benefit degrade the performance of those that do. Other developments in userlevel virtual memory management [18, 12, 7] and quality of service (QoS) 13, 3, 20] have provided the impetus for a more specialised and deterministic environment. My aim is to take advantage of the developments in QoS and user level memory management, combining these with solutions for the disk latency problem to provide a flexible, ....

McNamee, D., and Armstrong, K. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proceedings of the Worshop on Mach (Burlington, VM, USA, October 1990), USENIX Association, pp. 17--30.

....that enable application controlled caching policies. These include LRU SP for file caching with global guarantees [14] informed caching and prefetching for files using the access pattern hints from the application [94] and Mach VM page caching with application specified hints (PREMO pager) [82]. All these caching algorithms focus on minimizing response times for accesses predominantly on textual numeric and static image data. Since continuous media (CM) accesses are sequential, periodic and require real time guarantees, minimizing response times is not the primary goal. Consequently, ....

D. McNamee and K. Armstrong. Extending the Mach External Pager Interface to Accomodate User-level Page Replacement Policies. In Proceedings of Usenix Mach Workshop, pages 17--29, 1990.

....policies to user applications. Since the user applications know their access behaviors, when page faults or page replacements happen, the user applications can decide to return the least important page frames. Traditionally, such delegating mechanism is implemented in Upcall [2, 13] or IPC [18, 22] by transferring control from the operating system kernel to the user applications. These traditional domain crossing techniques creates evident overhead in transferring control and passing messages, which compromise system performance. In this paper, we challenge the necessity of domain crossing ....

....in making specific page frame management decisions, since direct accesses from user applications of the kernel data structures are prohibited. For example, when determining the dirty page frames to be flushed, the user applications need to invoke system calls to get the modified bits information [18]. Third, extra scheduling overhead is generated if the operating system has to reclaim page frames from a bundle of specific applications. Previous implementations [18, 22] extending page frame management policies created 10 to 14 overhead that follow the above observations. Unlike the ....

[Article contains additional citation context not shown here]

D. McNamee and K. Armstrong, 'Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies', Proceedings of the First USENIX Mach Workshop, Burlington, Vermont, USA, October 1990, pp. 17-29.

....no protection against rogue or buggy applications that crash the system or use the hardware to attack other systems. Other work spans file systems [Rees et al. 86, Bershad Pinkerton 88] scheduling [Anderson et al. 92] communication [Bershad et al. 91] and user level memory management [McNamee Armstrong 90, Harty Cheriton 92, Sechrest Park 91, Krueger et al. 93] The work in extensible filesystems permits applications to extend the semantics of files on a per file basis. However, this work still leaves all resource scheduling decisions to the operating system. With scheduler activations, the ....

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of the USENIX Association Mach Workshop, pages 17--29, 1990.

....of these techniques has ranged from architecture [46] to database systems [47, 11, 39, 13] to file systems [17, 33, 24, 37, 48, 6, 21, 9, 42] and beyond. A recent trend in this research is to use applications knowledge about their access patterns to perform more effective caching and prefetching [6, 9, 41, 42, 23, 34]. Our practical motivation for this problem comes from file systems. In this domain, the most common prefetching approach is to perform sequential read ahead, i.e. to detect when an application accesses a file sequentially, and to prefetch the blocks of the files that are so used [17, 33, 35] ....

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the First USENIX Mach Symposium, 1990.

....using the specific policy semantics. For example, a working set policy needs to maintain a window of pages that have been referenced and a page replacement algorithm based on that window. This particular separation of base meta functionality has been cogently argued in much existing literature[14, 33, 18]. Choices extends this basic meta hierarchy to two levels. An instance of the hierarchy for a Base Level LRU MRU WS 2nd Meta Level Level 1st Meta SO Pager S1 Sn Figure 2: Two level meta hierarchy in Choices w.r.t. cache replacement policy. virtual memory system is depicted in figure 2. The ....

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Usenix Mach Workshop, pages 17--29, October 1990.

....considerable control over the memory management system. However, there can still be problems. For example, Mach only allows control over what is done with a page once the OS decides to page it out. In this case, however, a solution is available: Mach has been extended by McNamee and Armstrong [141] to allow user level page replacement policies. In the next section, we discuss those dsm systems that support more than one hardware platform and those that support the automatic translation of data between hardware types. 3.6 Heterogeneity support A few of the dsm systems discussed make attempts ....

Dylan McNamee and Katherine Armstrong. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Mach Workshop, pages 17--29. USENIX Association, October 1990.

....the loop detector in that it tries hard to work well on applications where LRU is appropriate. The Atlas scheme apparently performed poorly for non scientific programs [9] Recent research projects on application controlled kernels show the potential of application specific replacement policies [28, 14, 21, 19]. These studies focus on mechanisms by which applications inform the kernel about what pages would be good candidates for replacements. Our SEQ algorithm is basically the antithesis of such schemes. It will be interesting to see over time which philosophy prevails. Our study shows that run time ....

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of USENIX Mach Symposiumi '91, pages 17--29, 1990.

.... systems have been modified to address performance problems caused by a particular application s needs, such as interprocess communication, synchronization, thread management, networking, virtual memory, and cache management [DBRD91, BRE92, SBC93, Ber93, YBMM94, MB93, Fel92, YTR 87, HC91, MA90, ABLL92, FP93, WB92, RLB94, ROKB95] For example, most improvements in IPC performance have been motivated by database applications or operating system servers. Each change required careful and deliberate modifications of the operating system kernel, making the extension difficult to implement. ....

....management policies that are accessible through interfaces exposed by the memory management system. Some of these interfaces have made it possible to manipulate large objects, for example entire address spaces [YTR 87, KN93] or to direct expensive operations, for example page7 out [HC91, MA90, CFL94] entirely from user level. Others have enabled control over relatively small objects, for example cache pages [RLB94] or TLB entries [BKW94] entirely from the kernel. None have allowed for fast, fine grained control over the physical and virtual memory resources required by applications. ....

Dylan McNamee and Katherine Armstrong. Extending the Mach External Pager Interface to Accommodate UserLevel Page Replacement Policies. In Proceedings of the USENIX Association Mach Workshop, pages 17--29, 1990.

....a mapped file. However, this interface is much more limited than ours, allowing specification of only a small number of basic patterns and no priorities. As a result, in the past few years there has been a stream of research papers on application control of caching in the virtual memory context [19, 28, 12, 16]. Our work differs from that described in these papers in three significant ways: ffl None of these papers (except [12] addresses the global allocation policy problem. By contrast, we discuss this problem in detail, and provide a solution, LRU SP, which we have simulated [3] and now have ....

....By contrast, we discuss this problem in detail, and provide a solution, LRU SP, which we have simulated [3] and now have implemented. ffl Most of these papers relied on RPC or upcalls for kernel to user communication, and consequently reported overhead as high as 10 of the total execution time [19, 28]. We provide a flexible interface for applications to issue primitives to exert control on cache replacement; we found that this is adequate most of the time and requires low overhead. ffl Most of these papers do not adequately consider which replacement policy an application should use. In [3] ....

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the USENIX Association Mach Workshop, pages 17-- 29, 1990.

....for User Level Virtual Memory Management Before describing our extensible virtual memory manager, we first outline the kernel support necessary to allow each application to set its own paging policy at user level. Support similar to what we describe here is provided by Mach (as extended by [McNamee Armstrong 1990], V , and Apertos. The key observation is to appropriately divide responsibility for virtual memory between the application, the kernel, and a separate user level paging system, specific to the application. This organization can be seen in Figure 1. In the simplest case, the application sees no ....

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of First USENIX Mach Symposium, Burlington, Vermont, October 1990.

....respect to file cache replacement. Simple schemes perform well, but do not give applications sufficient flexibility in controlling cache replacement. A more general scheme, for example, upcalls , can give applications complete control over cache replacement, but can incur high overhead [ABLL92, MA90] The main design challenge is to devise an interface that allows applications to exert the control they need, without introducing the overhead that would result from a totally general mechanism. The basic idea in ACFS s interface is to allow applications to assign priorities to their own files ....

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of USENIX Mach Symposiumi '91, pages 17--29, 1990.

....NORMA version of Mach, which implements network communication inside the micro kernel (the latency of communication on an 10 Mb Ethernet was 14 lowered from 50 ms for the Mach OSF 1 version to 4. 7 ms for the NORMA version) Lack of control over the page replacement algorithm As proposed in [McNamee 90] and [Krueger 93] it would be interesting to provide the ability to control which page must be flushed from the physical memory. For instance in Eliott, the cluster manager has some knowledge about cluster usage that the kernel does not have. In the current implementation, cluster managers ....

McNamee, D. and Armstrong, K., Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies, Proc. of the Mach Usenix Workshop, Burlington, VE, pp 31-43, Oct. 1990

....of resources other than CPU cycles only in a very coarse manner, which is typically based on static limits on total consumption. The development of more powerful policies to control the consumption of such resources has been the focus of complimentary research in application specific paging [27, 20, 24] and file caching [9] disk bandwidth allocation [46, 47] and TCP buffer management [39] 4.5 The resource container hierarchy Resource containers form a hierarchy. The resource usage of a child container is constrained by the scheduling parameters of its parent container. For example, if a ....

D. McNamee and K. Armstrong. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proc. USENIX Mach Symp., Oct. 1990.

....is implemented in a kernel extension model by allowing the set of processes that implement the multi threaded task to implement their own scheduling algorithm. The Mach external pager interface allows pagers to control how backing store is used, but not which pages to evict. McNamee and Armstrong [12] extended the model to allow external pagers to control page eviction. The level of granularity is still coarse; implementing a different paging algorithm implies implementing a new pager. We argue for a fine grained model that allows any particular algorithm to replaced in the context of an ....

McNamee, D., Armstrong, K. "Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies", USENIX Mach Workshop, pp. 17--29 (October 1990).

....is largely agreed that physical limitations will prevent any similar improvements in disk access time. To address this problem, work has been done to reduce the number of page faults by making the operating system more cognizant of each applications memory access patterns [Harty Cheriton 1992, McNamee Armstrong 1990, Krueger et al. 1993] Given this knowledge, a page replacement policy optimized for a particular application can yield significantly fewer faults. However, an operating system s default replacement policy very often minimizes the page faults which do occur. XMEM is not designed to lower the ....

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proceedings of First USENIX Mach Symposium, Burlington, Vermont, October 1990.

....prefetching) by any of the three policies and all hence all three policies perform equally well. 6 Related Work Many advanced operating systems and research prototypes have addressed the virtual memory page replacement problem. Mach [7] Spring [14] SPIN [8] HiPEC [15] V [10] and PREMO [6] all put the external memory management into their designs. The common technique is to have and application specific page replacement policy and downloaded into the kernel or have 2 The optimal strategy requires storing the entire traces in main memory to carry out a look ahead reference to ....

D. McNamee and K. Armstrong. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. Proceedings USENIX Association Mach Workshop, Burlington, Vermont, October, 1990.

....objects are associated with a pager. The pager is responsible for implementing the backing store for virtual memory. The initial pager design did not allow pagers to control eviction policy. McNamee and Armstrong extended the interface to allow user level control of page replacement policies [MCNAM90]. These extensions allow an application to specify the algorithm to use when selecting a page for eviction. ffl Vnode file system: The VFS model [KLEI86] abstracts the implementation details of a file systems, allowing the rest of a Unix 1 kernel to interact with it through a fixed interface of ....

....are not well served by the simplification. Typically, a DBMS knows in advance which pages will be needed soon, and could inform the operating system of its future requirements. There have been numerous attempts to empower applications with the ability to control prefetching and page replacement [CAO94, MCNAM90, PATT93, HART92, APPEL91]. All of these approaches provide potential solutions, but the solutions apply only to buffer management and require the implementation of special purpose mechanisms. In VINO, applicationspecific tailoring is the norm, not the exception. Every resource policy decision can be modified at the ....

McNamee, D., and K. Armstrong, "Extending The Mach External Pager Interface To Accommodate User-Level Page Replacement Policies," Proceedings of the 1990 Usenix Mach Workshop, Burlington, VT (1990).

....although it does allow application specified pagers. Unfortunately, the pager interface does not provide applications the ability to specify page replacement, it allows applications to specify only the backing store for memory pages. A new interface proposal provides this additional capability [19], but the solution is much larger than the problem. It is unclear that every application that cares about page replacement also cares about backing store. The granularity of service that can be specified in current operating systems is too large. Applications may need to replace just one routine ....

....to control the replacement policies for those regions. In the fine grain model, the replacement policy can be modified by specifying an application specific deallocate function. In the external pager model, McNamee and Armstrong had to substantially modify the existing external pager interface [19]. Instead of dictating which page to evict, the new interface allows the kernel to ask the external pager to evict a page of its choice. The kernel retains control over which memory object will lose a page frame, but the pager controls which page will be evicted. Unfortunately, this interface ....

McNamee, D., Armstrong, K., "Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies," Proceedings of the Usenix Mach Symposium, 1990, 17-- 29.

....we should know the capacity of a block and the internal memory beforehand, 2. we must be able to align elements into blocks, and 3. we must have a full control over the replacement of the blocks in internal memory. There are operating systems that provide support for these facilities (see, e.g. [17, 21, 23]) The tree structure of Arge and the heap structure of Fadel et al. do not give any guarantees for the performance of individual operations. In fact, one Insert or DeleteMin can be extremely expensive, the cost of handling the whole sequence being an upper bound. Therefore, it is risky to use ....

D. McNamee and K. Amstrong. Extending the Mach external pager interface to accommodate user-level block replacement policies. Technical Report 90-09-05, Department of Computer Science and Engineering, University of Washington, Seattle, 1990.

....and especially the management of the execution structures. Protected ports have a major drawback in a distributed environment: applications cannot share ports. Port group would have been convenient for managing distributed entities or providing fault tolerance facilities. As proposed in [McNamee 90] it would be interesting to provide the ability to manage page replacement in physical memory at the level of a memory managers, for the improvement of cluster paging. 4.3 Perspectives The Guide system is currently used for the development of new experimental applications, and as a platform ....

D. McNamee and K. Armstrong, Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies, Proc of the Mach Usenix Workshop, Burlington, VE, Oct 1990, pp 31-43

....kernel and the user process can choose from this menu. Examples of such replacement policies include: LRU with relative weights, MRU (most recently used) LRU K[21] etc. For full flexibility, the kernel can make an upcall to the manager process every time a replacement decision is needed, as in [18]. Similarly, each manager process can maintain a list of free blocks, and the kernel can take blocks off the list when it needs them. The manager would be awakened both periodically and when its free list falls below an agreed upon low water mark. This is similar to what is implemented in [25] ....

....a stream of research papers on mechanisms to implement virtual memory paging at user level. The external pager in Mach [29] and V [6] allows users to implement paging between local memory and secondary storage, but it does not allow users to control the page replacement policy. Several studies [18, 25, 12, 15] proposed extensions to the external pager or improved mechanisms to allow users to control page replacement policy. These schemes do not provide resource allocation policies that satisfy our design principles to guarantee replacement performance. Furthermore, they are not concerned with file ....

Dylan McNamee and Katherine Armstrong. Extending The Mach External Pager Interface To Accommodate User-Level Page Replacement Policies. In Proceedings of the USENIX Association Mach Workshop, pages 17--29, 1990.

No context found.

Dylan McNamee and Katherine Armstrong. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of the First USENIX Mach Symposium, pages 17--29, Burlington, Vermont, October 1990.

No context found.

D. McNamee and K. Armstrong, "Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies," in Proceedings of the First USENIX Mach Symposium, pp. 17--29, (Burlington, Vermont), October 1990.

.... system services, including interprocess communication, synchronization, thread management, networking, virtual memory, and cache management [Draves et al. 91, Bershad et al. 92, Stodolsky et al. 93, Bershad 93, Yuhara et al. 94, Maeda Bershad 93, Thekkath et al. 93, Felten 92, Young 89, McNamee Armstrong 90, Anderson et al. 92, Wheeler Bershad 92] In each case, the interfaces exported by a service were poorly matched to the needs of important applications. The solution to the performance problem came from enabling applications to adapt the behavior (interface and implementation) of system ....

....Kernel case II Application Server Kernel Application Server Kernel Figure 2: Comparison of a traditional microkernel server with a partitioned service in SPIN. 2. 4 An example As an example of how components of SPIN fit together, consider the structure of a user level virtual memory manager [McNamee Armstrong 90, Harty Cheriton 92] A user level virtual memory manager enables an application to control the set and contents of physical page frames that are currently backing a given piece of virtual memory. An application can request, say, 100 pages of physical memory from the system s physical page ....

[Article contains additional citation context not shown here]

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of the Usenix Mach Symposium, pages 17--29, 1990.

No context found.

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accommodate User-Level Page Replacement Policies. In Proceedings of the Usenix Mach Symposium, pages 17--29, 1990.

No context found.

D. McNamee and K. Armstrong. Extending the mach external pager interface to accomodate user-level page replacement policies. Technical Report 90-09-05, University of Washington Dept. of Comp. Sc. and Engg., 1990.

No context found.

McNamee, D. and Armstrong, K. Extending the Mach External Pager Interface to Accomodate User-Level Page Replacement Policies. In Proceedings of First USENIX Mach Symposium, Burlington, Vermont, October 1990.

No context found.

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. Technical Report TR-90-09-05, Department of Computer Science and Engineering, University of Washington, 1990.

No context found.

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. Technical Report TR-90-09-05, Department of Computer Science and Engineering, University of Washington, 1990.

No context found.

Dylan McNamee and Katherine Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. Technical Report TR-90-09-05, Department of Computer Science and Engineering, University of Washington, 1990.

No context found.

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the First USENIX Mach Symposium, 1990.

No context found.

D. McNamee and K. Armstrong. Extending the Mach external pager interface to accommodate user-level page replacement policies. In Proceedings of the First USENIX Mach Symposium, 1990.

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