Ramakrishnan, K., Biswas, P., and R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments," Proceedings of the Conference on Measurement and Modeling of Computer Systems, 1992, pp. 78-90.  Home/Search   Document Not in Database   Summary   ACM   TOC   Related Articles   Check

....workloads. In the case of system level simulation modeling, this consists largely of system level traces. Because of the popularity of the trace driven storage subsystem simulation, many companies include storage request trace acquisition instrumentation in their operating system software [Rama92, Ruemmler93, Treiber94]. Instrumentation for collecting system level traces, as described in section A.2.2, should also be included. With such instrumentation, large libraries of system level traces can be collected from real user environments. The instrumentation described in section A.2.2 limits the length of ....

....ffl The format of the disk request traces captured by the UNIX SVR4 MP device driver instrumentation described below. ffl The format of the extensive disk request traces from HP UX systems described in [Ruemmler93] ffl The format of the I O traces from commercial VMS systems described in [Rama92]. ffl A simple ASCII format, used mainly for debug purposes. Internal Storage I O Request Format During execution, the simulator reads request information from the trace file and translates it to the internal request format. The simulator handles differences between the endianness of the ....

[Article contains additional citation context not shown here]

K. Ramakrishnan, P. Biswas, R. Karelda, "Analysis of File I/O Traces in Commercial Computing Environments", ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, 1992, pp. 78--90.

....Increase Factor Figure 2.2: Failure induced workload increase in RAID Level 5. The figure shows the factor by which the surviving disk workload increases when a disk fails in a RAID Level 5 array [Ng92a] Note that the y axis starts at 1.0. 13 Since typical OLTP workloads are read dominated [Ramakrishnan92], the load on the surviving disks increases by typically between 50 100 in the presence of a disk failure. This severely degrades the performance as observed by the users, and dramatically lengthens the period of time required to recover the lost data and store it on a replacement drive. This ....

....of data can lead to significant disruption in the service provided by the application, we chose to model an on line transaction processing (OLTP) workload. OLTP workloads are characterized by a large number of independent processes, concurrently reading and writing data in relatively small units [TPCA89, Menon92c, Ramakrishnan92]. They typically mandate a minimum level of responsiveness; for example, the TPC A transaction processing benchmark [TPCA89] requires that 90 of all transactions complete in under two seconds. In many such environments, the system must service ad hoc database queries simultaneously with the ....

[Article contains additional citation context not shown here]

K. Ramakrishnan, P. Biswas, and R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments," Proceedings of the Conference on Measurement and Modeling of Computer Systems, 1992, pp. 78-90. 198

....parallel applications. General purpose workstations. Uniprocessor file access patterns have been measured many times. Floyd and Ellis [3, 4] and Ousterhout et al. 5] measured isolated Unix workstations, and Baker et al. measured a distributed Unix system (Sprite) 6] Ramakrishnan et al. [7] studied access patterns in a commercial computing environment on a VAX VMS platform. These studies all cover general purpose (engineering and office) workloads with uniprocessor applications. These studies identify several characteristics that are common among uniprocessor file system ....

K. K. Ramakrishnan, P. Biswas, and Ramakrishna Karedla, "Analysis of file I/O traces in commercial computing environments", in Proceedings of ACM SIGMETRICS and PERFORMANCE '92, 1992, pp. 78--90.

....sector variance] 8KB mean request size [exponential] 0 22 ms interarrival [uniform] 4.2 Workloads We use synthetically generated random workloads and extensive traces captured from actual systems. We will describe the traces only briefly as they have been described elsewhere in more detail [Rama92, Ruem93]. The traced workloads span a broad range of environments, and each trace is at least a full workshift (8 hours) in length. Two of the traces come from Hewlett Packard systems running HP UX TM , a version of the UNIX TM operating system [Ruem93] Cello comes from a server at HP Labs used for ....

....University of California, Berkeley used primarily for compilation and editing. While these traces are actually two months in length, we report data for a single week long snapshot (5 30 92 to 6 6 92) The other four traces are from commercial VAX TM systems running the VMS TM operating system [Rama92]. Air Rsv is from a transaction processing environment in which approximately 500 travel agents made airline and hotel reservations. Sci TS is from a scientific time sharing environment in which analytic modeling software and graphical and statistical packages were used. Order and Report are from ....

[Article contains additional citation context not shown here]

K. Ramakrishnan, P. Biswas, R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments ", ACM SIGMETRICS, 1992, pp. 78-90.

....characteristics of the disk request traces. 2.2 Disk Request Traces For our experiments, we use six traces of disk activity collected from systems in use at various industrial and research installations. We describe the traces only briefly as they have been described elsewhere in more detail [Ramakrishnan92, Ruemmler93]. The traced workloads span a range of user environments, and each is at least a full workshift (8 hours) in length. Some basic characteristics of the traces are given in table 1. They vary widely in read write ratios, access sizes, arrival rates, locality and burstiness. Two of the traces come ....

....of California at Berkeley used primarily for compilation and editing. While these traces are actually two months in length, we report data for a single week long snapshot (5 30 92 to 6 6 92) The other four traces are from commercial VAX TM systems running the VMS TM operating system [Ramakrishnan92]. Air Rsv is from a transaction processing environment in which approximately 500 travel agents made airline and hotel reservations. Sci TS is from a scientific time sharing environment in which analytic modeling software and graphical and statistical packages were used. Order and Report are from ....

K. Ramakrishnan, P. Biswas, R. Karelda, "Analysis of File I/O Traces in Commercial Computing Environments", ACM SIGMETRICS Conference, 1992, pp. 78--90.

....figure observed over all validation runs was 1.9 of the corresponding average response time. 4.2 Workloads We use synthetically generated random workloads and extensive traces captured from actual systems. We describe the traces only briefly as they have been described elsewhere in more detail [Ruem93, Rama92]. The traced workloads span a broad range of environments, and each is at least a full workshift (8 hours) in length. Two of the traces come from Hewlett Packard systems running HP UX TM , a version of the UNIX TM operating system [Ruem93] Cello comes from a server at HP Labs used for program ....

....of California at Berkeley used primarily for compilation and editing. While these traces are actually two months in length, we report data for a single week long snapshot (5 30 92 to 6 6 92) The other four traces are from commercial VAX TM systems running the VMS TM operating system [Rama92]. Air Rsv is from a transaction processing environment in which about 500 travel agents made airline and hotel reservations. Sci TS is from a scientific time sharing environment in which analytic modeling software and graphical and statistical packages were used. Order and Report are from a ....

K. Ramakrishnan, P. Biswas, R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments", ACM SIGMETRICS, 1992, pp. 78-90.

....total workload, typically deriving from decision support or array maintenance functions rather than actual transactions. Thus, for OLTP environments, only a small minority of the user accesses touch more than one data unit, and the fraction of reads that access more than G 1 units is even smaller [Ramakrishnan92]. Therefore the benefits of achieving criterion six in our layout would be marginal at best in OLTP workloads. We will not considered further the problem of meeting this criterion 4 . P P P P P D3 D8 D13 D2 D7 D12 D17 D1 D6 D11 D16 D0 D5 D10 D15 DISK0 DISK1 DISK2 DISK3 DISK4 0 1 2 3 Offset ....

....[Katz89] It consists of four primary components, illustrated in Figure 7. The top level of abstraction contains a synthetic reference generator. Table 2a shows the workload generated for our simulations. This workload is based on access statistics measured on an airline reservation OLTP system [Ramakrishnan92]. The requests produced by this workload generator are sent to a RAID striping driver, whose function is to translate each user request into the corresponding set of disk accesses. Table 2b shows the configuration of our extended version of this striping driver. Low level disk operations generated ....

K. Ramakrishnan, P. Biswas, and R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments," Proceedings of the Conference on Measurement and Modeling of Computer Systems, 1992, pp. 78-90.

....when transactions that do not run concurrently access the same data. Because caching strives to retain data at a site even after a transaction has completed, the cache consistency maintenance algorithm must also deal effectively with this type of sharing. Recent studies of file system workloads [Ramakrishnan et al. 1992; Sandhu and Zhou 1992] indicate that sequential sharing may, in fact, be quite common in the types of situations in which page servers are intended to be used. Two approaches to reducing the potential for aborts in optimistic techniques have been proposed. One is to treat hot spot data ....

Ramakrishnan, K. K., Biswas, P., and Karedla, R. 1992. Analysis of file I/O traces in commercial computing environments. In Proc. 1992 ACM SIGMETRICS and PERFORMANCE '92 Int'l. Conf. on Measurement and Modeling of Computer Systems (Newport, Rhode Island, USA, June 1-5 1992), pp. 78--90.

....which is capable of producing user request streams drawn from a variety of distributions. Table 5 1 (a) shows the configuration of the workload generator used in our simulations. We have restricted our attention to random accesses of size 4 KB to model an OLTP system with an effective buffer cache [Ramakrishnan92]. Each request produced by this generator is sent to a RAID striping driver, which was originally the actual code used by the Sprite operating system [Ousterhout88] to implement a RAID device on a set of independent disks. Table 5 1 (b) shows the configuration of our extended version of this ....

K. Ramakrishnan, P. Biswas, and R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments, " Proceedings of the ACM SIGMETRICS Conference, 1992, pp. 78-90.

....when transactions that do not run concurrently access the same data. Because caching strives to retain data at a site even after a transaction has completed, the cache consistency maintenance algorithm must also deal effectively with this type of sharing. Recent studies of file system workloads [Rama92, Sand92] indicate that sequential sharing may, in fact, be quite common in the types of situations in which page servers are intended to be used. If this is the case, then the naive use of optimistic techniques could result in unacceptably high abort rates. Two approaches to reducing the potential for ....

Ramakrishnan, K., Biswas, P., Karedla, R., "Analysis of File I/O Traces in Commercial Computing Environments", Proceedings of the ACM SIGMETRICS and Performance '92 Conference, May, 1992.

.... Baker et al. examined changes in file access patterns six years later, and also investigated issues of file sharing in the Sprite distributed operating system [1] Ramakrishnan et al. performed a similar analysis of file system traces collected from large customers of Digital Equipment Corporation [15]. File system traces and simulated file system workloads have also been used to as input for file system simulations and to stress test new file system architectures. Both simulated workloads and file system traces have been used by different researchers to evaluate garbage collection strategies ....

Ramakrishnan, K., Biswas, P., Karedla, R. "Analysis of File I/O Traces in Commercial Computing Environments." Proc. 1992 SIGMETRICS. Newport, RI. June 1992. pp. 78--90.

....used as input to controlled experimental investigations. For this work the focus is on requests for file service by applications running on client workstations. Several previous studies have addressed the characterization of requests for file service. Results reported in [2] 10] 12] 13] and [14] suggest that file references exhibit properties of temporal and spatial locality similar to those found in memory reference traces. In other words, the same tendencies for references to information items to be clustered spatially (spatial locality) and to persist over substantial time periods ....

Ramakrishnan, K.K., Biswas, P. and Karedla, R., "Analysis of File I/O Traces in Commercial Computing Environments", Proc. 1992 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, Newport, Rhode Island, June 1992, 78-90.

....model the components of the I O subsystem of interest: the disk controller, the read and write caches which we postulate are part of the controller, and the magnetic media. The traces we use for the simulation are detailed I O operation traces from several commercial time sharing VAX VMS systems [10]. The traces are quite large (between 100 500 K operations per disk) and each of the environments studied had a large number of users. All the I O operations to the disk subsystem are captured with detailed time stamps and the logical block number of the disk blocks accessed. We also had available ....

....in all the environments) and the users were unaware of the tracing. Therefore, we believe that the traces capture activity from truly operational, commercial production systems. A detailed description of the tracing mechanism, the traces and the characteristics of file I O activity can be found in [2, 10]. Based on the environment and file access behavior observed, the traces were broadly categorized into three major interactive classes: Office and Scientific Time Sharing . Time Sharing with some database activity . Transaction processing We have chosen one trace from each of these three ....

[Article contains additional citation context not shown here]

Ramakrishnan, K.K., et al, "Analysis of File I/O Traces in Commercial Computing Environments, " Proceedings of the 1992 ACM SIGMETRICS and PERFORMANCE '92 International Conference on Measurement and Modeling of Computer Systems, June 1992.

No context found.

Ramakrishnan, K., Biswas, P., and R. Karedla, "Analysis of File I/O Traces in Commercial Computing Environments," Proceedings of the Conference on Measurement and Modeling of Computer Systems, 1992, pp. 78-90.

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