R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation, " in Proc. 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....prediction of the program s performance. Unfortunately, this is also the drawback of simulators: they need to do substantial work for every dynamic instruction. There are many simulators currently being used in academics and industry, a sample are SimOS [RHW 95] SimICS [MDG 98] Talisman [B95], SimpleScalar [BA97] RSIM [PRA97] FastSim [SL98] and perft5 [T97] There are two main classes of simulators: emulators and trace based simulators. An emulator actually executes the instructions in the program. This allows software to be tested on the simulator and also instills some ....

R. Bedichek. Talisman: Fast and Accurate Multicomputer Simulation. Performance Evaluation Review, vol. 23, no. 1, pp. 14-24, May 1995.

....it uses a functional first decoupling to implement a static full system simulator. In another, it pioneers the timing directed organization. Conversely, several previously published simulators are dynamic full system simulators, but are not decoupled. g88 [4] gsim [19] and Talisman [5] are in order functional simulators for the 88000 processor, which can simulate a modified version of UNIX. SimOS is a dynamic full system simulator that supports out of order processor models for the MIPS [23] and Alpha instruction sets [3] PharmSim is a dynamic fullsystem simulator based on ....

....recovery does not affect memory Name Full System Out of Order Multiprocessor Multiscalar Simulator [6] Yes Yes FastSim [24] Yes Yes SimpleScalar 3. 0 [2] Yes Yes MASE [18] Yes Yes RSIM [15] Yes Yes SimpleMP [22] Yes Yes Yes ASIM [14] Yes Yes Yes g88 [4] Yes gsim [19] Yes Talisman [5] Yes Simics [20] Yes Yes PharmSim [7] Yes SimOS [23] Yes TFsim [This paper] Yes Yes Yes Yes indicates OS modifications are required. Table 1. Summary of related simulators values, as no speculative data leaves the processor (similar to real systems) However, the recovery changes ....

R. C. Bedichek. Talisman: Fast and accurate multicomputer simulation. In Proceedings of the 1995.

....devices. This makes it possible to run an entire operating system on top of the simulator, and include OS activity in the evaluations, as well as run unmodified programs. The inclusion of OS or system software activity is important, as it can have a significant impact [3] This approach is used in [2] [10] 11] A weakness is that such a simulator tend to be computationally expensive, and even if efficient instruction set simulators exist today, it is not always practical to construct one. The construction of such a simulator is often extremely complicated, due to the need to recreate the ....

R. C. Bedichek, "Talisman: Fast and Accurate Multicomputer Simulation", in Proceedings of the `95 ACM SIGMETRICS Conference, pp. 14-24, 1995.

....interaction (ii) Device emulation (a) Cost accuracy tradeoff (b) Trace driven simulation vs device emulation Figure 2: Performance evaluation techniques and tradeoffs. tion in real systems rapidly become intractable. Another technique is (trace driven) simulation, which has several advantages [5]. Since a simulator is built in software, it can be readily modified and augmented to test new features and interfaces. Simulators are usually easier and cheaper to build than real systems. They can model ideal systems that are impossible to build, e.g. an infinitely fast network. However, a ....

....itself. Simulation based evaluation: Performance evaluation via simulation can be conducted at various levels of detail, and hence, accuracy. Recently, significant attention has been given to accurate, low level simulation to study machine architectures while capturing operating system overheads [5, 21]. Other efforts have focused on protocol level simulation with the ability to run the actual protocol stack during simulation [6] and network level simulation with a focus on routing and end to end protocol performance [22, 23] Most relevant to our work is architecture level and protocol level ....

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation," in Proceedings of Sigmetrics 95/Performance 95, pp. 14--24, May 1995.

....behavior of network traffic. However, such models are usually simplified to make the analysis tractable, and they rarely account for system overheads of interrupt handling, context switches, etc. encountered in practice. Another technique is simulation, which has several significant advantages [2], including ease of development, flexibility and versatility. Since a simulator is built in software, it can be readily modified and augmented to test new features and interfaces. Simulators are usually easier to build and cheaper than real systems. They can model ideal systems that are ....

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation," in Proceedings of Sigmetrics 95/Performance 95, pp. 14--24, May 1995.

....two protocols are identical in instruction and data cache behavior. Although we refer to FlashLite as the memory system simulator, it actually models everything in the FLASH system other than the main microprocessor and its caches. SimOS. The SimOS simulation environment [19] like the Talisman [1] and SimICS simulators [10] models the system in enough detail to boot and run a full operating system. We are able to run a slightly modified version of IRIX 6.4 on our simulated machine. The kernel modifications are mostly to the boot code and I O device drivers due to dif ferences between ....

R. Bedichek. Talisman: Fast and Accurate Multicomputer Simulation. Performance Evaluation Review, vol. 23, no. 1, pp. 14-24, May 1995.

....requiring user assistance. Dynamo s input is an already compiled native instruction stream, that is re optimized to exploit performance opportunities that manifest themselves at runtime. A lot of work has been done on dynamic translation as a technique for non native system emulation [8] 30][5][31] 12] 17] The idea is to lower emulation overhead by caching native code translations of frequently interpreted regions. Unlike such binary translators, Dynamo is not concerned with translation. The Dynamo approach does however allow one to couple a fast lightweight translator that emits ....

Bedichek, R. 1995. Talisman: fast and accurate multicomputer simulation. In Proceedings of the 1995 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems.

....however, our prototype currently bails out to direct program execution if a fork or exec is encountered in its code. Related Work A lot of work has been done on dynamic translation as a technique for non native system emulation [May 1987; Insignia 1991; Cmelik and Keppel 1993; Stears 1994; Bedichek 1995; Witchel and Rosenblum 1996; Hohensee et al. 1996; Herold 1998] The basic idea is simple: interpretation is too expensive a way to emulate a long running program on a host machine. Caching the native code translations of frequently interpreted regions can lower the overhead. If sufficient time ....

Bedichek, R. 1995. Talisman: fast and accurate multicomputer simulation. In Proceeding 1995 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems.

....limiting them to toy benchmarks or very patient users. This has prompted several efforts to improve the performance of traditional simulation or to find alternate methods. This work has met with some success: several fast instruction set simulators have been developed over the last several years (Bedichek 1990 and 1995, Cmelik and Keppel 1994, Goldschmidt 1993, Rosenblum et al. 1995, Veenstra and Fowler 1994, Witchel and Rosenblum 1996) 1.2 Instruction Caches An execution profile showing the frequency of execution of any line of code is a traditional component in performance tuning, and will be meaningful to ....

Bedichek, R. C. 1995. Talisman: Fast and accurate multicomputer simulation. In Proceedings of the '95 SIGMETRICS Conference, 14-24.

....beyond these eight processors on the grid is performed by message passing which is transparent to the user. Message passing is achieved through memory to memory copy. The four ports to memory are implemented as four caches connected on the same bus with the memory module as depicted in Figure 2. The processor memory cluster consists of two basic boards; namely, the processor board and the memory board. The processor board has two versions, to implement the desired connectivity and allow easy expansion. The memory board contains four memory modules. The lines at the end of the split ....

....a diffusion algorithm which ensures that the workload is evenly balanced among the processors [13] 3. MODELLING PiSMA Several tools have been developed for the modeling and simulation of parallel computer structures; SIMON [ 7] ADAS [18] PAWS [16] Network II.5 [ 5] FRIGG [ 3] Talisman [ 2] are some examples of such tools. However, although these generic tools exhibit a number of useful features, they do not often provide the functionality required to match the specific needs and objectives of the designer of a particular architecture. Thus, it is often a typical case in computer ....

[Article contains additional citation context not shown here]

Bedichek, R., "Talisman: Fast and Accurate Multicomputer Simulation", SIGMETRICS 1995.

....of network traffic. However, such models are usually simplified to make the analysis tractable, and they rarely account for system overheads of interrupt handling, context switches, etc. encountered in practice. Another feasible technique is simulation, which has several significant advantages [5]. Since a simulator is built in software, it can be readily modified and augmented to test new features and interfaces. Simulators are usually easier to build and cheaper than real systems. They can model ideal systems that are impossible to build, e.g. an infinitely fast network. However, a ....

....evaluation via simulation can be conducted at various levels of detail, and hence, accuracy. Recently, significant attention has been given to accurate, low level simulation to study machine architectures while capturing operating system overheads typically encountered in real systems [5, 12]. Other efforts have focused on protocol level simulation with the ability to run the actual protocol stack during simulation [6] and network level simulation with a focus on routing and end to end protocol performance [13, 14] Most relevant to our work is architecture level and protocol level ....

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation," in Proceedings of Sigmetrics 95/Performance 95, pp. 14--24, May 1995.

....systems, but to point out directions in which the simulation of parallel computers is developing. Hardware Simulation Simulation can be done at different levels of abstraction, resulting in different levels of achievable accuracy. One of the low level hardware simulators is Talisman (Bedichek [3]) It reaches quite high accuracy at the cost of 100 host instructions per simulated instruction. Hardware Emulation An alternative way to speed up model execution is to use hardware emulation instead of simulation. Barroso et al. 2] present their emulator Rpm, which speeds up model execution by ....

R. C. Bedichek, Talisman: Fast and Accurate Multicomputer Simulation, Performance Evaluation Rev. 23--1, special issue (1995), pp. 14--24.

....results are presented in section 6. In Section 7, the simulation performance of Mermaid is discussed. Finally, section 8 concludes the paper and mentions possible future work. 2 Simulation methodology In the last few years, many multiprocessor simulation systems have been proposed and implemented [1, 17, 4, 6, 12, 19, 3, 8, 18]. Many of these systems use execution driven simulation and apply a technique called direct execution. In this technique, two types of instructions are distinguished: local and non local instructions. An instruction is local if its execution affects only the local processor. For example, all ....

R. C. Bedichek. Talisman: Fast and accurate multicomputer simulation. In Proceedings of the 1995 ACM SIGMETRICS Conference, pages 14--24, May 1995.

....used the first 1,000,000 references of the SPEC92 gcc trace. For the same experiment, the quad processor server s throughput improved by 400 over the non threaded version of TSF. With 40,000,000 references to process, TSF had been able to simulate 74,000 references per second. The Bedichek [1] simulator used threading and on a SPARC 10 server was able to simulate 850,000 cycles of execution in 6 minutes which is equivalent to 2400 references per second assuming one reference per cycle. System observations have shown that TSF is very efficient as the majority of available compute power ....

R Bedichek. Talisman: Fast and accurate multicomputer simulation. SIGMETRICS, pages 14--24, 1995.

....simulation can be extended to perform a full logical to physical translation on every memory access and simulating memory mapped devices, allowing the interpreter to support arbitrary programs including multitasking operating systems running on parallel hardware. An example is Talisman [7]. Talisman performs a detailed simulation of the Meerkat [6] memory system, doing a function call on each memory access. This scheme slows down the simulation, but permits accurate timing. A speedier predecessor to Talisman, g88 [5] simulated an infinite TLB using a double indirection that was ....

R. C. Bedichek. Talisman: Fast and Accurate Multicomputer Simulator. In Proceedings of the '95 ACM SIGMETRICS Conference, 1995. Personal Communication (to be published).

....and negative individual values from canceling each other out. Note the average error is under 5 percent, with the worst case error being about 15 percent. We are aware of only a very few pieces of work that use trace driven or execution driven simulation that actually validate their simulators [10, 25, 36]. Our accuracy is comparable to theirs. More details about the construction and validation of the simulator can be found in the appendix. 4.4 Characterization and Analysis In this Section, we present our characterization and analysis of memory reference behavior of network protocols under a ....

....goal of this simulator has been to accurately model performance costs for our SGI machine. Much of the simulation literature discusses the tradeoff between speed and accuracy, and describes techniques for making simulations fast. However, accuracy is rarely discussed (notable exceptions include [10, 25, 36]) and the tradeoff between accuracy and speed has not been quantitatively evaluated. Given that our simulator is meant to capture performance costs, it must be more than an emulator that duplicates the execution semantics of the hardware or counts events such as cache misses. The simulator ....

Bedichek, R. C. Talisman: Fast and accurate multicomputer simulation. In Proceedings of the ACM Sigmetrics Conference on Measurement and Modeling of Computer Systems, pages 14--24, Ottawa, Canada, May 1995.

....policies to application performance requirements [13 17] Evaluating and tuning such router designs requires special support in the network simulator. Several recent simulation tools evaluate various aspects of multicomputer applications and interconnection networks. Executiondriven simulators [18 20] typically capture the instruction level operation of applications on particular multicomputer architectures. Other simulation tools emphasize multicomputer network architectures, allowing users to vary the router s buffer architecture, switching scheme, and routing algorithm, under different ....

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation," in Proc. ACM SIGMETRICS /Performance, pp. 14--24, May 1995.

....are major concerns in simulator design and implementation. Many articles have studied the issues of efficiency and accuracy of different simulation methods (e.g. 7] and presented examples in simulating important components of a computer system, and specific computer architectures (e.g. [3, 6]) We emphasize the flexibility of simulating the heterogeneity at different system levels that may exist in a NOW System. We give a simulation framework to integrate other existing uniprocessor, memory and network simulators. We address the issues of the efficiency, accuracy and generality of our ....

R. C. Bedichek, "Talisman: fast and accurate multicomputer simulation", Proceedings of the 1995 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May 1995.

....355 0x00011400 27070 0x00011500 16685 0x00011600 160 0x00011700 15239915 . Figure 11: Linux boot statistics (gdb simics) devs [ 1] 19714] from 0x08400000 to 0x08400ffe is the dma [ 2] 12484] from 0x08800000 to 0x08800ffe is the esp [ 3] 7] from 0x10000000 to 0x10002ffe is the iommu [ 4] [ 0] from 0x50200000 to 0x50200ffe is the display [ 5] 0] from 0x50300000 to 0x50300ffe is the display [ 6] 0] from 0x50700000 to 0x50700ffe is the display [ 7] 0] from 0x50800000 to 0x508fdffe is the display [ 8] 79] from 0xd0000000 to 0xd0000ffe is the zs [ 9] 389] from 0xd0001000 to ....

....one at a time. This provides an accessible and in some sense correct target machine model for instrumentation, and places minimum restrictions on the architectural relationship between the host and target. Examples include g88 [3] CacheMire [7] Mint [28] Shade [8] SimOS [26, 30] Talisman [4], and SimICS. All these simulators translate from a target code to an intermediate format. This format can then either be interpreted [3, 4, 19] or directly executed [2, 20, 26, 28, 30] Instruction set simulation is generally the slowest but most flexible approach. 8.6 SimICS SimICS is an ....

[Article contains additional citation context not shown here]

R. C. Bedichek. Talisman: Fast and Accurate Multicomputer Simulation. In Proceedings of the '95 SIGMETRICS Conference, 1995.

....of such an emulator is SPIM, which reads and translates a MIPS I executable, in its entirety, to an intermediate representation understood by the emulation engine [Larus91] After translation, SPIM can lookup and emulate predecoded instructions with a slowdown factor of approximately 25. Talisman [Bedichek95] and gsim [Magnusson93] also use a form of instruction predecoding, but instead of decoding all instructions of a workload before it begins running, these emulators predecode instructions lazily, as they are executed for the first time. By caching the results, these emulators can benefit from ....

Bedichek, R. Talisman: fast and accurate multicomputer simulation. In Proceedings of the 1995 SIGMETRICS Conference on Measurement and Modeling of Computer Systems, 14-24, 1995.

No context found.

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation, " in Proc. 1995.

No context found.

Bedichek R.: Talisman: Fast and Accurate Multicomputer Simulation. In \SIGMETRICS '95", ACM, NewYork 1995., pp. 14-24.

No context found.

BEDICHEK, R. "Talisman: fast and accurate multicomputer simulation". Proceedings of SIGMETRICS'95, Ottawa, Ontario, Canada. May 1995.

No context found.

R. C. Bedichek, "Talisman: Fast and accurate multicomputer simulation," in Proceedings of 1995 ACM SIGMETRICS Conference, pp. 14--24, 1995.

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