E. Riedel and G. Gibson, "Understanding Customer Dissatisfaction with Underutilized - 28 - Distributed File Servers," Proceedings of the 5th NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... capability requests maximum load a given percentage active minutes with capability traffic is higher maximum load same percentage without capability traffic, then server will need more resources promptly service that percentage client minutes if managers wish minimize customer dissatisfaction [Riedel96]. 5.1.3.1 Sample Workloads I studied two workloads: a Berkeley NFS trace CMU AFS 1999 traces. The Berkeley NFS trace records activity Auspex fileserver supporting client Filemanager NASD Client Figure 5 1 Capability Requests Impact Latency sequence of small requests, such a recursive find a ....

....system. NASD needs support a wide range workloads from filesystems databases, with wide range access patterns. examine current systems, diverse set access patterns. For example, academic research software development group, typical system accesses small while most data moved large requests [Baker91, Riedel96]. addition file systems, databases and persistent object systems operate small objects [Stamos84] These results both small access large transfers are important consider for NASD. Section discussed support for large read accesses using stored message digest, small accesses pose different issues. ....

Riedel, E., Gibson, "Understanding Customer Dissatisfaction with Underutilized Distributed File Servers," Proceedings Fifth NASA Goddard Space Flight Center Conference Mass Storage Systems Technologies. College Park, MD. September

....node , with its own CPU and memory in a disk controller. DR net al..so provides fault tolerance to multiple disk failures. A great deal of work related to function distribution in a disk system has been reported. In [1] 5] 10] Active Disk, a successor of Network attached secure disks (NASD) 6][9], is discussed. This approach aims to shift part of the application programs to disk drives and to reduce the data transfer between an application server and disk drives. IDISK[7] has a general purpose CPU, a memory capacity of tens of megabytes and gigabit communication links in the disk drives. ....

E. Riedel and G. Gibson. Understanding customer dissatisfaction with underutilized distributed file servers. In Proc. of the 5th NASA Goddard Space Flight Center Conf. on Mass Storage Systems and Technologies, 1996.

....file systems in the public domain. In this section we present a comparison study of DFS with two other existing file systems. We use the popular network file system (NFS) 18] that is widely used in a distributed file sharing environment. We also use CMUNASD and the file system associated with it [16], 10] henceforth referred to as NASD FS. Also note that it is difficult, if not impossible, to make a fair comparison among these file systems as will be apparent below. Network File System: The Network File System (NFS) was designed by Sun Microsystems in 1985 to allow transparent file sharing ....

E. Riedel and G. A. Gibson. Understanding Customer Dissatisfaction With Underutilized Distributed File Servers. In Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996.

....In this section we present a comparison study of DFS with two other existing le systems. We use the popular network le system (NFS) 23] that is widely used in a distributed le sharing environment. The other le system we use for the comparison is the NASD and the le system associated with it [21, 12], henceforth referred to as NASD FS. Also note that it is dicult, if not impossible, to make a fair comparison among these le systems as will be apparent below. Network File System: The Network File System (NFS) was designed by Sun Microsystems in 1985 to allow transparent le sharing among ....

....It resides on devices separate from those that contain data, so accessing the meta data does not interfere with the streaming of data. 6. 3 Network Attached Storage Devices The improvements in the networking and storage technologies can be exploited by attaching storage directly to the network [21]. Distributed le systems using network attached storage have two major performance advantages. First, the simple, expensive, and data intensive operations are o loaded from the le server machines and thus more number of clients can be supported by the le servers. Second, by coupling le ....

[Article contains additional citation context not shown here]

E. Riedel and G. A. Gibson. Understanding Customer Dissatisfaction With Underutilized Distributed File Servers. In Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996.

....In this section we present a comparison study of DFS with two other existing le systems. We use the popular network le system (NFS) 16] that is widely used in a distributed le sharing environment. The other le system we use for the comparison is the NASD and the le system associated with it [14, 9], henceforth referred to as NASD FS. Also note that it is dicult, if not impossible, to make a fair comparison among these le systems as will be apparent below. 19 Network File System: The Network File System (NFS) was designed by Sun Microsystems in 1985 to allow transparent le sharing among ....

E. Riedel and G. A. Gibson. Understanding Customer Dissatisfaction With Underutilized Distributed File Servers. In Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996.

....local disks, they are accessed locally. The local disk cane be used to buffer data retrieved from the network attached RAID. The TickerTAIP [7] Network attached Secure Disks [13] 14] active disks [34] and Storage Area Network [39] are good examples of network attached storage system [24] 30][35]. This architecture provides direct data transfer between central storage and clients in a networking environment. This architecture is more scalable than the use a central storage server by direct network striping [6] Figure 2 shows the system architecture of a network attached RAID subsystem. ....

E. Riedel, and G. A. Gibson, "Understanding Customer Dissatisfaction With Underutilized Distributed File Servers", Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, Sept. 1996.

.... percentage of the active minutes with capability traffic is higher than maximum load for the same percentage without capability traffic, then the server will need more resources to promptly service that percentage of the client minutes if the managers wish to minimize customer dissatisfaction [Riedel96]. 5.1.3.1 Sample Workloads I studied two workloads: a Berkeley NFS trace and CMU AFS 1999 traces. The Berkeley NFS trace records activity on an Auspex fileserver supporting 231 client Filemanager NASD Client Figure 5 1 Capability Requests Impact on Latency For a sequence of small requests, such ....

....from filesystems to databases, with a wide range of access patterns. If we examine current systems, we see a diverse set of access patterns. For example, in an academic research software development group, typical file system accesses are small while most data is moved in large requests [Baker91, Riedel96]. In addition to file systems, databases and persistent object systems operate on small objects [Stamos84] These results tell us that both small access and large transfers are important to consider for NASD. Section 6.4 discussed support for large read accesses using stored message digest, but ....

Riedel, E., and Gibson, G., "Understanding Customer Dissatisfaction with Underutilized Distributed File Servers," Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies. College Park, MD. September 1996.

....provide higher availability, especially for the site disaster tolerance. Distributed RAID in cluster can provide parallel I O functionality. This is very important to the I O intensive applications, such as multimedia applications and I O 2 intensive massive parallel computing applications [3][19]. More and more researches on parallel I O are carried on a virtual shared memory on top of a distributed shared memory (DSM) Workstation cluster offers a single I O space on top of distributedmemory hardware architecture. Research on distributed RAID in clusters can provide through ....

E. Riedel and G. A. Gibson, "Understanding Customer Dissatisfaction With Underutilized Distributed File Servers", Proceedings of the Fifth NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996

....to better share peripherals and take advantage of the relative technological trends of buses, networks, and peripheral processors. The architectural reasons to shift from host adapterattached devices to network attached are better sharing of devices and reduction of the server s workload [RG96] By allowing clients to directly access the devices, the server is no longer in the data path, reducing latency and demands on its buses, memory and processors. Devices can also communicate directly with each other without sending data across a single shared system bus. Buses do not scale well. ....

....with other types of networks. The CMU Parallel Data Lab s Network Attached Secure Disk (NASD) project divides NFS like functionality between a file manager and the disk drives themselves, so that not only read and write commands but also attribute set and get are executed at the drive [G 96, RG96, G 97] The file manager is responsible primarily for verifying credentials and establishing access tokens. Soltis Global File System (GFS) uses Fibre Channel disk drives modified to support a lock primitive [SRO96, Sol97] This provides simple, efficient distributed locking. The drive ....

[Article contains additional citation context not shown here]

Erik Riedel and Garth Gibson. Understanding customer dissatisfaction with underutilized distributed file servers. In Kobler [Kob96], pages 371--388. also known as CMU-CS-96-158.

....of magnitude slower than a server cache access. As a result, the average time to access a file block increases considerably, limiting performance even though the hit rate to the local client caches might be high[Dahlin94] Such a phenomenon has been observed even when the server is lightly loaded[Riedel96]. There are two principal reasons for the poor hit rate on the server cache: capacity and locality. First, as the number of clients increases, the number of server accesses increases proportionately. However, the server cache is fixed in size and does not have the capacity to satisfy the ....

E. Riedel and G. Gibson. Understanding Customer Dissatisfaction with Underutilized Distributed File Servers. In Proceedings of the 5th NASA Goddard Space Flight Center Conferenceon Mass Storage Systems and Technologies, September 1996.

No context found.

E. Riedel and G. Gibson, "Understanding Customer Dissatisfaction with Underutilized Distributed File Servers," Proc. Fifth NASA Goddard Space Flight Center Conf. on Mass Storage Systems and Technologies, 1996, http://www.pdl.cs.cmu.edu/PDLFTP /NASD/Goddard96.abstract.html.

....dataless clients [Ruemmler93] AFS 1994 217 GB 4,750 47 900 clients, 70 servers, server data only [Spasojevic96] HP 1994 54 GB 2. 3 million 527 105 4,363 46 machines [Sienknecht94] Harvard 1994 23 GB 75 314 four machines [Smith94] Carnegie Mellon 1996 8 GB 25 328 single server, server data only [Riedel96] Carnegie Mellon 1998 26.5 GB 40 671 single server, server data only Microsoft 1998 10.5 TB 140 million 4,418 2,492 31,689 4,800 machines [Douceur99] Table 2 9 Amount of storage used in an organization. The table compares the amount of data and the total number of users across several years of ....

Riedel, E. and Gibson, G. "Understanding Customer Dissatisfaction With Underutilized Distributed File Servers" Fifth Goddard Conference on Mass Storage Systems and Technologies, September 1996.

....NFS Cost Measurements Table 3(c) Table 3(d) 7 4.4 Busy Client Minutes A distributed file system scales if an increase in aggregate client demand, and the corresponding increase in storage capacity and bandwidth, does not result in a decrease in client observed performance. In a previous study [Riedel96], we examined the correlation between hourly averages of client response times, network round trip times and server load. Users may be satisfied with their response times when servers are idle, but experience periods of dramatically longer response times which correlate with periods of high server ....

Riedel, E. and Gibson, G. "Understanding Customer Dissatisfaction With Underutilized Distributed File Servers", 5th Goddard Conference on Mass Storage Systems and Technologies, College Park, MD, Sept. 1996. 13

No context found.

E. Riedel and G. Gibson, "Understanding Customer Dissatisfaction with Underutilized - 28 - Distributed File Servers," Proceedings of the 5th NASA Goddard Space Flight Center Conference on Mass Storage Systems and Technologies, September 1996.

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