C. P. Thacker and L. C. Stewart. Firefly: A Multiprocessor Workstation. In Proceedings of the Second International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS-II, pages 164--172, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....dataflow information [7, 21, 22, 23, 25, 30] With extra cost, multiprocessors may employ dedicated hardware for cache coherence maintenance by allowing processors to communicate with each other about the data reference status, and to invalidate or update cached copies. Snoopy buses [12, 15, 29, 35, 38] and memory directories [2, 4, 5, 14, 37] are two prominent hardware coherence mechanisms. With run time interproces2 sor dataflow information, the coherence hardware never over invalidates the cached data like the software schemes, and therefore generally outperforms the software schemes. ....

C. P. Thacker and L. C. Stewart. Firefly: A multiprocessor workstation. In Proc. Second International Conference on Architectural Support for Programming Languages and Operating Systems, pages 164--172, October 1987.

....faults, distort the equivalence between virtual and physical processors; in the presence of these factors, user level threads built on top of traditional processes can exhibit poor performance or even incorrect behavior. Multiprocessor operating systems such as Mach [Tevanian et al. 87] Topaz [Thacker et al. 88] and V [Cheriton 88] provide direct kernel support for multiple threads per address space. Programming with kernel threads avoids system integration problems: the threads that are used by the programmer or compiler are directly scheduled by the kernel. Unfortunately, the performance of thread ....

Thacker, C., Stewart, L., and Satterth- waite, Jr., ]. Firefly: A Multiprocessor Workstation. IEEE Transactions on Computers, 37(8):909-920, August 1988.

....access fault will occur. At. this point. the copying thread immediately copies the page and sets the access for the page to read write and restarts the faulting thread. Several benchmark programs have been used t. o measure the performance of this algorithm on the DEC Fire fly nmltiprocessors [33]. The measurements show that about 90 of the checkpoint work is executed concurrently with the target program while no thread is ever interrupted for more than .1 second at. a time. This method also applies to taking incremental checkpoints: saving the pages that have been changed since the ....

Charles Thacker, Lawrence Stewart, and Edwin Sat- terthwaite. Firefly: A multiprocessor workstation. IEEE Transactions on Computers, 37(8):909-920, Au- gust 1988.

....systems use inactive thread migration to balance load. To name just a few, the Topaz operating 23 1 1 2 P 1 P 3 P 0 P Figure 2 3. Message pattern under computation migration. A thread on processor P 0 makes n tra c is not shown. system [72] for the Firefly multiprocessor workstation [92] migrates inactive threads. Markatos [70] explored a scheduling policy that favors locality over load balance: threads are initially scheduled based on expected accesses, and idle threads are migrated to balance load. Anderson et al. 3]studied, among other things, the performance implications of ....

C.P. Thacker, L.C. Stewart, and E.H. Satterthwaite Jr. "Firefly: A Multiprocessor Workstation ". In Proceedings of the 2nd Conference on Architectural Support for Programming Languages and Systems, Palo Alto, CA, October 5--8, 1987.

....protocols have been specified in the literature. These include both invalidate based and update based protocols. The invalidate based protocols include Goodman s write once [37] the Synapse [30] the Illinois [58] and the Berkeley [29] protocols. The update based protocols include the Firefly [67] and the Dragon [54] protocols. These protocols have been well studied [46, 26, 8, 7, 54, 67] The next two sections give a brief description of how the Berkeley invalidatebased protocol and the Dragon update based broadcast protocols operate for typical processor reads and writes. Berkeley ....

....and update based protocols. The invalidate based protocols include Goodman s write once [37] the Synapse [30] the Illinois [58] and the Berkeley [29] protocols. The update based protocols include the Firefly [67] and the Dragon [54] protocols. These protocols have been well studied [46, 26, 8, 7, 54, 67]. The next two sections give a brief description of how the Berkeley invalidatebased protocol and the Dragon update based broadcast protocols operate for typical processor reads and writes. Berkeley Invalidate Based Protocol In the Berkeley invalidate based protocol, a cache line may be in one ....

Charles P. Thacker, Lawrence C. Stewart, and Edwin H. Satterthwaite Jr. Firefly: A Multiprocessor Workstation. In Proceedings of the Second International Conference on Architectural Support of Programming Languages and Operatings Systems, pages 164--172, October 1987.

....sending the same data over a 10Mb s Ethernet, the high speed transmission will complete before the Ethernet does, despite the high latency. Remote Procedure Call [Birrell Nelson] or RPC) has received considerable attention in the effort to reduce latency. A new version of RPC for the Firefly [Thacker, et al. 88] multiprocessor which breaks down the time used for each of 15 different steps in an RPC operation can be found in [Schroeder Burrows 89] The objective was that to understand how to reduce the latency of an RPC operation, one must discover the source of most of the latency and center ....

C. P. Thacker, L. C. Stewart & E. H. Satterhwaite Jr. Firefly: A Multiprocessor Workstation. IEEE Transactions on Computers, August 1988, pp. 909-920.

.... Illinois [Papamarcos84] Berkeley [Katz85] RB (Read Broadcast) Rudolph84] and EIP (Efficient Invalidation Protocol) Archibald87] The Write Update (WU) class consists of those protocols that update remote copies upon a write operation involving a shared copy: Dragon [McCreight84] Firefly [Thacker88], and RST (Reduced State Transition) Prete91] A first evaluation of most of these protocols can be found in [Archibald86] Two new WU protocols have been defined for two special bus based machines: on chip multiprocessor [Takahashi96] and bus based COMA [Lee94] A first attempt to standardize ....

C. Thacker, L. Stewart, and E. Satterthwaite, "Firefly: a multiprocessor workstation," IEEE Trans. Computers, vol. 37, no. 8, pp. 909--920, Aug. 1988.

....the written data is broadcast on the bus to all other caches, which then update their copies of the line. Cache invalidations are never performed by the cache consistency protocol. Two examples are the protocol in DEC s Firefly multiprocessor workstation and that in the Xerox Dragon multiprocessor [TS87, AM87]. Protocols in this class are called write broadcast protocols. Each of these two classes of protocol has certain advantages and disadvantages, depending on the pattern of references to the shared data. For a shared data line that tends to be read and written several times in succession by a ....

CHARLES P. THACKER AND LAWRENCE C. STEWART. "Firefly: a Multiprocessor Workstation". Proceedings Second International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS II), Palo Alto, California, IEEE Computer Society Press, October 5--8, 1987, pages 164--172.

....short. Even though kernel threads running in the same user process share the same address space, they are scheduled independently by the kernel scheduler. Each scheduling requires the transfer of control between the kernel level and the user level, still causing too much overhead. Mach[3, 4] Topaz[1, 9], and V[10] support the kernel thread mechanism. The other one is the user level thread. User level threads are managed by runtime routines pre compiled and linked to the user application as a library. User level execution is transparent to the kernel. Scheduling among the userlevel threads does ....

C. Thacker, L. Stewart, and E. Satterthwaite, Jr., "Firefly: A Multiprocessor Workstation, " in IEEE Trans. Comput., vol 37, no. 8, pp. 909-920, Aug, 1988.

....58 Firefly Firefly is a shared memory multiprocessor workstation of DEC Systems Research Center that contains from one to seven processors. The processors have coherent caches, so that each sees a consistent view of main memory. See Topaz for description of the software system. Reference: [363] Galaxy Galaxy is a distributed operating system designed at the university of Tokyo. It provides network wide shared memory access. Reference: 364] GFS The Generic File System for UNIX has been created by Digital Equipment Corporation to support all types of hierarchical file systems ....

C.P. Thacker and L.C. Stewart, "Firefly: A Multiprocessor Workstation", In Int. Conf. on Architectural Support for Programming Languages and Operating Systems, pages 164--172, Palo Alto, Ca., February 1987.

....how data is moved among the caches in the machine, ensuring that data is frequently found in the local cache, while preventing processors from using stale data. There are two common classes of coherence protocol used in shared memory machines: writeupdate protocols (WU) McCreight, 1984; Thacker and Stewart, 1987; Thacker et al. 1992] and write invalidate protocols (WI) Goodman, 1983; Papamarcos and Patel, 1984; Lenoski et al. 1990] Under a WU protocol, each time a processor writes shared data, the coherence protocol broadcasts the new value to every other processor caching that data. Under a WI ....

....an update for a block that is only cached by the updating processor, the acknowledgement of the update instructs the processor to retain future updates since the data is effectively private. This optimization is analogous to the shared line on the bus of the Dragon [McCreight, 1984] and Firefly [Thacker and Stewart, 1987] multiprocessors. Second, when a parallel process is created by fork, we flush the cache of the parent s processor, which eliminates useless updates of data initialized by the parent but not subsequently needed by that processor. 2.2 Performance Metrics The focus of this paper is on our ....

Charles P. Thacker and Lawrence C. Stewart, "Firefly: a Multiprocessor Workstation," In Proceedings of the 2nd International Conference on Architectual Support for Programming Languages and Operating Systems, pages 164--172, Palo Alto, CA, October 1987.

....miss, the line comes from its owner or from shared memory. All other caches invalidate their copy if any. The line is loaded in state M . Similar principles to those discussed above in the framework of Write Invalidate protocols also apply to Write Update protocols, such as the Firefly protocol [Thacker et al. 88] In fact, quite a number of both types of snoopy coherence protocols have been proposed and implemented (e.g. Goodman 84, Fielland et al. 84, Rudolph et al. 84, Katz et al. 85, Archibald 86, Archibald et al. 86, Goodman 87] These range from the simple Write Through protocol [Fielland et al. 84, ....

THACKER, C.P., STEWART, L.C., AND SATTERTHWAITE, E.H. Firefly : A Multiprocessor Workstation. IEEE Transactions on Computers, pp.909--920, Vol.37, No.8, August 1988.

....bus (see Figure 2.37) The sources and sinks in these models represent the extra load due to the cache coherence protocol. These models have been studied with Write Once [Goodman 84] Write through [Fielland et al. 84] Berkeley [Katzetal85] Synapse [Frank 84] Illinois [Patel et al. 84] Firefly [Thacker et al. 87] and Dragon [McCreight 84] cache coherence protocols. As an example, the results for the Write Once protocol (using the workload parameters fr =0:7, fw =0:3, h =0:9, md =0:2, umd =0:1, Nsb =32, qs =0:2, Sc =2K, Z =2, tcc =1, tcb =1 and tcm =4) are shown in the Figure 2.35. As we shall see in ....

THACKER,C.,AND STEWART, L. Firefly: A Multiprocessor Workstation. Computer Architecture News, pp.164--172, Vol.15, No.5, October 1987.

.... have introduced concurrency are MultiprocessorSmalltalk (MS) 14] 15] and ConcurrentSmalltalk (CST) 21] MS is an example of Smalltalk where processes (units of execution) execute in parallel like HoME, and it is a system which implements Berkeley Smalltalk [19] on the FireFly multiprocessor [17] with the V distributed operating system [4] Berkeley Smalltalk is a type which unlike HPS interprets byte code. In MS, concurrent execution is realized by replicating the interpreter and having the same number of interpreters as the number of physical processors. The ready queue for processes is ....

Charles P. Thacker and Lawrence C. Stewart, Firefly: a multiprocessor workstation, Proc. of APSLOS II, Computer Society Press of the IEEE, 164172, October 1987.

....access fault will occur. At this point the copying thread immediately copies the page and sets the access for the page to read write, and restarts the faulting thread. Several benchmark programs have been used to measure the performance of this algorithm on the DEC Firefly multiprocessors [33]. The measurements show that about 90 of the checkpoint work is executed concurrently with the target program while no thread is ever interrupted for more than .1 second at a time. This method also applies to taking incremental checkpoints; saving the pages that have been changed since the last ....

Charles Thacker, Lawrence Stewart, and Edwin Satterthwaite. Firefly: A multiprocessor workstation. IEEE Transactions on Computers, 37(8):909--920, August 1988.

....I O, and page faults, distorts the equivalence between virtual and physical processors; in the presence of these factors, user level threads built on top of traditional processes can exhibit poor performance or even incorrect behavior. Multiprocessor operating systems such as Mach [2] Topaz [22], and V [7] provide direct kernel support for multiple threads per address space. Programming with kernel threads avoids the system integration problems exhibited by user level threads, because the kernel directly schedules each application s threads onto physical processors. Unfortunately, kernel ....

....space thread scheduler on a kernel event; the thread scheduler can use the activation to modify user level thread data structures, to execute user level threads, and to make requests of the kernel. We have implemented a prototype of our design on the DEC SRC Firefly multiprocessor workstation [22]. While the differences between scheduler activations and kernel threads are crucial, the similarities are great enough that the kernel portion of our implementation required only relatively straightforward modifications to the kernel threads of Topaz, the native operating system on the Firefly. ....

THACKER, C., STEWART, L., AND SATTERTHWAITE, JR., E. Firefly: A multiprocessor workstation. IEEE Trans. Comput. 37, 8 (Aug. 1988), 909--920.

....several server based systems are being implemented. These systems are characterized by a small kernel or micro kernel that provides only basic communication services, and multiple servers providing all other services. Examples of such systems include Chorus, 7 Topaz, and Version 3 of Mach [1, 18, 22]. Server based systems attempt to find a balance between modularity and performance by supporting coarse grained protection and modularity, with the granularity at the level of the protection domain. Each service is allowed to define its own interface. The interfaces are implemented on top of a ....

C. P. Thacker, L. C. Stewart, and E. H. Satterthwaite. Firefly: A multiprocessor workstation. IEEE Transactions on Computers, 37(8):909--920, Aug. 1988.

....memory. When a processor detects a memory reference to an object that has been copied into its local memory, it either invalidates [Goo83,Kat85] its local copy so that the next reference will force a current copy to be obtained from global shared memory, or it updates the copy with the new value [Atk87, Tha87]. The term snooping usually refers to this type of coherence mechanism for busbased, shared memory multiprocessors, but the same principles can be applied to maintain coherence between I O and cache, between cache and the SMC, between different FIFOs in the SMC, or even between I O and the SMC. ....

....This splitting of the synchronization operation into two separate phases allows an even greater overlap of memory operations by all processors. 8.1. 2 Coherence Mechanisms The coherence mechanisms that implement these memory consistency models fall into two general categories: snooping schemes [Arc86,Goo83,Kat85,Tha87], and directory based schemes [Aga88,Cen78,Cha91,Len90,OKr90] The best solution for a given system Chapter 8: Other Systems Issues 159 depends on several factors, including the number of processors, the anticipated workloads, the desired memory consistency model, and the desired system cost. ....

C.P. Thacker and L.C. Stewart, "Firefly: A Multiprocessor Workstation", Proceedings of the Second International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS-II), published as ACM SIGARCH Computer Architecture News, 15(5):164-172, October 1987.

No context found.

C. P. Thacker and L. C. Stewart. Firefly: A Multiprocessor Workstation. In Proceedings of the Second International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS-II, pages 164--172, 1987.

No context found.

C. P. Thacker, L. Stewart, and E. H. Satterthwaite. Firefly: A multiprocessor workstation. IEEE Transactions on Computers, 37(8):909--920, Aug. 1988. 11

No context found.

C. P. Thacker, L. C. Stewart and E. H. Satterthwaite, `Firefly: a multiprocessor workstation', IEEE Trans. Computers, C-37, (8), 909-920 (1988).

No context found.

C. P. Thacker, L. C. Stewart and E. H. Satterthwaite, `Firefly: a multiprocessor workstation', IEEE Trans. Computers, C-37, (8), 909--920 (1988).

No context found.

C. P. Thacker and L. C. Stewart, `Firefly: a multiprocessor workstation', Proceedings of the Second International Conference on Architectural Support for Programming Language and Operating Systems, 1987, pp. 164--172.

No context found.

C. P. Thacker, L. C. Stewart and E. H. Satterthwaite, `Firefly: a multiprocessor workstation', IEEE Trans. Computers, 37, (8), 909--920 (1988).

No context found.

C. P. Thacker, L. C. Stewart and E. H. Satterthwaite, `Firefly: a multiprocessor workstation', IEEE Trans. Computers, 37(8), 909--920 (1988).

First 50 documents  Next 50

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