Ann L. Chervenak and Randy H. Katz. Performance measurement of the first RAID prototype. Technical Report UCB/CSD 90/574, Computer Science Division (EECS), University of California, Berkeley, May 1990.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....3.1.2 Disk array model The disk array simulator was a simple array of first order disk simulators. No attempt was made to model interference among disks such as contention for I O bandwidth at a shared junction. Although real life experience has proven that such interference can be a bottleneck [Chervenak90], we assumed that these bottlenecks would be removed by hardware designers. The individual disk simulators included a request queue, an elevator scheduling algorithm (taken from the HP UX file system) and a model of the disk hardware. This hardware model had analytical models for seek time as a ....

Ann L. Chervenak. Performance measurements of the first RAID prototype. UCB/CSD 90/574. University of California at Berkeley, May 1990.

....components: a) 1 Sun 4 280 with 128MB of memory. b) 7 32 bit VME SCSI Host Bus Adapters. c) 32 CDC Wren IV 340MB 5.25 disks. d) 1 Ethernet Interface. The Berkeley RAID group discovered a hierarchy of bottlenecks which limited the overall performance of their system [E.K. Lee 90, Chervenak 90, Chervenak et al. 91, Chervenak 91] Performance was critically affected by a pathological interaction between the Sun 4 280 s memory system and the Sprite operating system. These problems were addressed in their second generation RAID II prototype [E.K.Lee 91] and strongly influenced the ....

CHERVENAK, A.L. Performance Measurements of the First RAID Prototype. Technical Report UCB/CSD 90/674, University of California, May 1990.

.... distributed computing and transaction processing applications, requires very frequent but small data accesses [91] However, measurements on the first RAID prototype at the University of California, Berkeley revealed poor performance and less than expected linear speedup for large data transfers [35]. The excessive memory copying overhead due to interaction of caching and DMA transfers, and restricted I O interconnect (VME bus) bandwidth were cited to be the primary reasons of poor performance. Also, it is recognized now that large RAID disk arrays do not scale very well in terms of ....

Chervenak, A, "Performance Measurements of the First RAID Prototype," Technical Report, Department of Computer Science, University of California, Berkeley, 1990.

....[Kro86] are two examples of SDA systems. 2.3. CSA and SDA Advantages and Drawbacks Previous studies show that under certain workloads the server in the CSA does not scale well because memory and I O bus contention during I O operations are serious bottlenecks that limit overall throughput [Che88], Tie94] Thus, to support larger numbers of clients, a CSA must either graduate to faster, more expensive servers, or to increase the number of servers. Again, the former solution is often preferred, since multiple servers introduce their own drawbacks. In particular, system administration, ....

....access protocols (CS, EWS, and SIO) described in section 4. The simulation was written in C using the CSIM library [Sch91] a process oriented, discrete event simulation package. Its performance was validated by comparing the results of the CSA simulation to the benchmarks from a previous study [Che88]. The logical model of the simulation is presented in Figure 4. Simulation parameters are given in Appendix A. The following subsections describe the logical components modeled by the simulation. 6.1. Client Process The simulated clients randomly pick a file identifier based on the access ....

Chervenak, A., 'Performance Measurements of the First RAID Prototype', University of Calif., Berkeley, Tech. Report UCB/CSD 90/574.

....architecture [6] a server computer provides a set of services to the client systems linked to it by a communication network. The CSA does not scale well because memory contention during I O operations, and I O bus contention on the server are serious bottlenecks which limit overall throughput [7]. Also, the CSA has low availability since server failures prevent client access to shared data. In contrast, the serverless shareddisk architecture (SDA) 8] allows every client direct access to the networked storage. However, in the context of interactive multimedia client coordination ....

Chervenak, A. Performance measurements of the first raid prototype. Tech. Rep. UCB/CSD 90/574, University of California, Berkeley, 1988.

.... distributed computing and transaction processing applications, requires very frequent but small data accesses [20] However, measurements on the first raid prototype at the University of California, Berkeley revealed poor performance and less than expected linear speedup for large data transfers [7]. The excessive memory copying overhead due to interaction of caching and dma transfers, and restricted i o interconnect (vme bus) bandwidth were cited to be the primary reasons of poor performance. Also, it is recognized now that large raid disk arrays do not scale very well in terms of ....

Chervenak, A, "Performance Measurements of the First RAID Prototype," Technical Report, Department of Computer Science, University of California, Berkeley, 1990. Large Scale Multimedia Servers 17

....A configuration is the total collection of variable value pairs used in an experiment. The base configuration is the collection of variable value pairs given in the tables in this section. Table 5 shows the base configuration variables for the hardware. The values for this table were taken from [2]. The disks and I O buses are simulated as follows. Requests for a disk read arrive from the CPU (after determining that they are cache misses) Each read has a specified length in bytes. The reads are queued at the disk in a first come first served (FCFS) manner. Each request is first serviced by ....

Ann L. Chervenak. Performance measurements of the first raid prototype. Technical Report UCB/UCD 90/574, University of California, Berkley, May 1990.

....hosts with a CPU and memory, several I O buses per host and several disks per I O bus. The hosts are connected by a local area network. See Table 5 for the parameters and base values that describe the the hardware configuration. The values for the disk and I O bus portions of this table are from [4]. The hosts have parameter values which correspond to a typical workstation. See Figure 1 for an example hardware configuration. 4.2 Inverted Lists and Answer Sets In our simulation, we do not generate a synthetic document base a priori. Instead, when we require the length of the inverted list ....

Ann L. Chervenak. Performance measurements of the first raid prototype. Technical Report UCB/UCD 90/574, University of California, Berkley, May 1990.

....a CPU and memory, several I O buses per host and several disks per I O bus. The hosts are connected by a local area network. See Table 5 for the parameters and base values that describe the the hardware configuration. The values for the disk and I O bus portions of this table are from reference [Che90]. The hosts have parameter values that correspond to a typical workstation. Figure 1 shows an example hardware configuration. CHAPTER 3. DISTRIBUTED QUERIES ANALYTIC WORKLOAD 34 Parameter Value Description CPUSpeed 1 Relative speed of each CPU Multiprogram 4 Multiprogramming level per Host ....

....is the total collection of variable value pairs used in an experiment. The base configuration is the collection of variable value pairs given in the tables in this section. Table 13 shows the base configuration variables for the hardware. The values for this table were taken from reference [Che90]. These values are better than, say, the theoretical maximum performance values available from the manufacturer s literature. The disks and I O buses are simulated in the same way as Chapter 3. Requests for a disk read arrive from the CPU (after determining that they are cache misses) Each ....

Ann L. Chervenak. Performance measurements of the first RAID prototype. Technical Report UCB/UCD 90/574, University of California, Berkeley, May 1990.

....A configuration is the total collection of variable value pairs used in an experiment. The base configuration is the collection of variable value pairs given in the tables in this section. Table 4 shows the base configuration variables for the hardware. The values for this table were taken from [2]. The disks and I O buses are simulated as follows. Parameter Value Description DiskBandwidth 10.4 Mbits sec bandwidth per disk DiskBuff 32768 Size of a disk buffer in bytes BlockSize 512 Bytes per disk block SeekTime 15.0 Disk seek time in ms T rackToT rack 4.0 Cost to seek one track in ....

A. L. Chervenak. Performance measurements of the first raid prototype. Technical Report UCB/UCD 90/574, University of California, Berkley, May 1990.

....query algorithms, we implemented a detailed event driven simulation using the DeNet [13] simulation environment. For details of the simulation, see [ The parameters controlling the hardware organization are listed in Table 5. The values for the disk and I O bus portions of this table are from [4]. In our simulation, we do not generate a synthetic document base a priori. Instead, when we require the length of the inverted list for a word w, we use the Parameter Value Description DiskBandwidth 10.4 Mbits sec Bandwidth DiskBuff 32768 Size of a disk buffer BlockSize 512 Bytes per disk ....

A. L. Chervenak. Performancemeasurements of the first raid prototype. Technical Report UCB/UCD 90/574, University of California, Berkley, May 1990.

.... distributed computing and transaction processing applications, requires very frequent but small data accesses [22] However, measurements on the first raid prototype at the University of California, Berkeley revealed poor performance and less than expected linear speedup for large data transfers [6]. The excessive memory copying overhead due to interaction of caching and dma transfers, and restricted i o interconnect (vme bus) bandwidth were cited to be the primary reasons of poor performance. Also, it is recognized now that large raid disk arrays do not scale very well in terms of ....

Chervenak, A, "Performance Measurements of the First RAID Prototype," Technical Report, Department of Computer Science, University of California, Berkeley, 1990. Large Scale Multimedia Servers 25

.... distributed computing and transaction processing applications, requires very frequent but small data accesses [24] However, measurements on the first raid prototype at the University of California, Berkeley revealed poor performance and less than expected linear speedup for large data transfers [12]. The excessive memory copying overhead due to interaction of caching and dma transfers, and restricted i o interconnect (vme bus) bandwidth were cited to be the primary reasons of poor performance. Also, it is recognized now that large raid disk arrays do not scale very well in terms of ....

Chervenak, A., "Performance Measurements of the First RAID Prototype," Technical Report, Department of Computer Science, University of California, Berkeley, 1990.

....and Experiences RAID I, the first RAID prototype, consisted of a SUN 4 280 file server with 128 MB of memory, four Interphase dual SCSI channel host bus adapters, and 32 340 MB 5. 25 Wren IV disk drives (4 drives on each of 8 SCSI channels) While we had hoped the design would be disk limited, Chervenak 91] discovered numerous performance bottlenecks. The most serious is the server s memory system, which limited application throughput to only 2.3 MB s. I O operations caused excessive memory to memory copies and cache flushes. The array did better on small random reads, achieving nearly 300 per ....

A. L. Chervenak, R. H. Katz, "Performance Measurements of the First RAID Prototype," Proceedings ACM SIGMETRICS Conference, San Diego, CA, (May 1991).

....at U.C. Berkeley recently built a prototype block interleaved disk array. This paper examines the performance limits of each component of the array using Sprite operating system traces, user programs and SCSI bus traces. Performance of software for implementing redundancy is considered elsewhere [Cher90,Lee90,Lee91]. We wanted to learn whether a disk array built from off the shelf parts could deliver adequate performance for two different workloads. A disk array has many independent disk arms and can perform many small, random I Os in parallel. Such I Os are typically generated by database applications and ....

....on power up. Trace analysis shows that data bytes do pass between the devices at 250 nsec per byte (4 MBytes sec) The sustained transfer rate is limited to 3 MBytes sec by overheads associated with SCSI protocol implementation on the disk and controller. SCSI overheads are described in detail in [Cher90]. Table 2 lists the percentage of time and the normalized time per I O spent in each phase of the SCSI protocol for a trace of 32 KByte sequential reads. A separate process issued these reads to each of four disks on a single SCSI string. The time spent in the arbitration phase is very short, ....

Ann L. Chervenak, "Performance Measurements of the First RAID Prototype", U.C. Berkeley Technical Report UCB/CSD 90/574, May, 1990.

No context found.

Ann L. Chervenak and Randy H. Katz. Performance measurement of the first RAID prototype. Technical Report UCB/CSD 90/574, Computer Science Division (EECS), University of California, Berkeley, May 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