M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in the system. This is typically done in one of four ways: increasing the capacity of the system for example by using server farms or multiprocessor machines [20, 22] using caches either on the client or on the server side [26, 15, 13] designing more efficient software both at the OS level [34, 7, 23, 29] and the application level [37] and admission control [17, 44] Other means of avoiding overload are content adaptation [1] and offloading work to the client [3] Our work differs significantly from all these approaches in that we do not attempt to reduce the load on the overloaded device. ....

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.

....full advantage of a cache of mmap( ed les the way an single process event driven server such as Zeus does. In contrast to the research above, which attempts to reduce the use of operating system services, other work has centered around improving the OS performance directly: Kaashoek et al. [21, 22] advocate a customized operating system tailored speci cally for servers. They demonstrate a prototype HTTP server OS that they claim performs an order of magnitude better than a conventional OS. Mechanisms they utilize include a uni ed disk bu er cache and network retransmission bu er, an ....

M. Frans Kaashoek, Dawson Engler, Gregory R. Ganger, and Deborah A. Wallach. Server operating systems. In 1996.

.... literature focuses on reducing network latency, e.g. by caching requests ( 21] 12] 11] or improving the HTTP protocol ( 19] 32] Other literature works on reducing the delays at a server, e.g. by building more efficient HTTP servers ( 20] 33] or improving the server s OS ( 18] 5] [26], 29] Recent studies show that delays at the server make up a significant portion of the response time [8] 7] Our work focuses on reducing delay at the server by using sizebased connection scheduling. In the remainder of this section we discuss only work on priority based or size based ....

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop, September 1996.

.... literature focuses on reducing network latency, e.g. by caching requests ( 21] 12] 11] or improving the HTTP protocol ( 19] 32] Other literature works on reducing the delays at a server, e.g. by building more efficient HTTP servers ( 20] 33] or improving the server s OS ( 18] 5] [26], 29] Recent studies show that delays at the server make up a significant portion of the response time [8] 7] Our work focuses on reducing delay at the server by using sizebased connection scheduling. In the remainder of this section we discuss only work on priority based or size based ....

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop, September 1996.

....applications handling of the packet can be contained separately to the actual application) Improvements of 80 to 100 are observed over a popular Unix clone. Other observed application level benefits include a 5 to 10 fold improvement in throughput performance of a WWW server on the Exokernel [Kaashoek,96] compared to commercial equivalents and similar results for such a server over SPIN. Finally, application domain specific improvements are reported [Bershad,94] for parallel processing, multimedia and database applications over their implementation on conventional operating systems. 153 7.3.6 ....

Kaashoek, M.F., Engler, D.R., Ganger, G.R., Wallach, D.A., "Server Operating Systems". Proceedings of the 7th SIGOPS European Workshop, Connemara, County Galway, Ireland, September 1996.

....can be divided into 3 primary areas. One solution is to increase the server resources, for example, adding more hardware, bandwidth, or CPU. This is often done by replacing the server by a server farm (e.g. 8, 6, 16] Another solution is to enhance the OS to better support server software: [20, 9, 1, 15, 19]) Finally, people try to limit the load at a server, either via installing a proxy cache (at the client or server end) e.g. 10, 4] or by admission control ( 13] Our approach to coping with transient overload is different from the above approaches. Our approach does not require buying more ....

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.

....implementing each state is directly linked with others in the flow of execution. Nonblocking I O is provided by most OSs, but these interfaces typically do not scale well as the number of I O flows grows very large [2, 14, 18] Much prior work has investigated scalable I O primitives for servers [2, 5, 13, 16, 22, 23, 25], but these solutions are often an afterthought lashed onto a process based model, and do not always perform well. To demonstrate this fact, we have measured the performance of the nonblocking socket interface in Linux using the dev poll [23] event delivery mechanism, which is known to scale ....

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach. Server operating systems. In Proceedings of the 1996 SIGOPS European Workshop, September 1996.

....both on disk and in memory. Even though the file system API is similar to Hummingbird, the Bullet service does not perform clustering of files together, so it would not have the same type of performance improvement that Hummingbird has for a caching web proxy workload. Kaashoek et al. [9] approaches high performance through developing server operating systems, where a server operating system is a set of abstractions and runtime support for specialized, high performance server applications. Their implementation of the Cheetah web server is similar to Hummingbird in one way: ....

Kaashoek, M. F., Engler, D. R., Ganger, G. R., and Wallach, D. A. Server operating systems. In Proceedings of the 1996 SIGOPS European Workshop (Connemara, Ireland, Sept. 1996), pp. 141--148.

....computation. Previous webserver research has shown that OS support for specialization of the web server, including avoiding checksum calculations through precomputation, can improve web server throughput by a factor of 2. 3 over the Harvest httpd accelerator acting as front end to an NCSA server [Kaashoek96]. This suggests that a similar optimization may be useful for storage systems. However, there is a problem applying this optimization to a storage system. In web applications, documents are requested in their entirety while storage requests are frequently for portions of a file. It is impractical ....

....MAC All Data MAC Normal MAC Inline Partial MAC 140 6.7 Related Work The Hash and MAC approach borrows its basic idea from the web server community. Kaashoek et al. demonstrated that precomputation of TCP IP checksums on web pages can help increase web server throughput but a factor of 2. 3 [Kaashoek96]. The stored hash values in NASD exploit workload similarities between web servers and fileservers. However, the Hash and MAC construction also handles the following concerns: any stored work must be usable for multiple clients, a MAC must be bound to a key, and clients can make requests on ....

Kaashoek, M. F., Engler, D. R., Ganger, G. R., and Wallach, D. A., "Server Operating Systems," 1996 SIGOPS European Workshop. Connemara, Ireland, 1996.

....through the system to be managed as a unit. We intend to apply similar techniques in SEDA to provide more complex load conditioning algorithms. Much prior work has investigated scalable I O primitives for server applications. We intend to build upon mechanisms for scalable network and disk I O [2, 22, 43, 33, 9] and I O event delivery [3, 27, 34, 35] incorporating these primitives into implementations of SEDA. Sandstorm s asynchronous sockets layer makes use of the dev poll event delivery mechanism as first described by [3] The vast body of literature in scheduling algorithms is relevant to the ....

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach. Server operating systems. In Proceedings of the 1996 SIGOPS European Workshop, September 1996.

....a resource principal. Consequently, it does not provide e ective support for di erentiated QoS in server environments where signi cant amounts of processing associated with a service class might be done in a separate address space (e.g. for CGI programs) 5.1. 9 Exokernels The exokernel approach [67, 41, 68] advocates moving bulk of the operating system software into libraries that can be customized speci cally according to application needs. The actual kernel is kept as small as possible and principally multiplexes access to the hardware safely between the hosted applications. Applications can exert ....

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach. Server Operating Systems. In Proceedings of the Seventh ACM SIGOPS European Workshop, Connemara, Ireland, Sept. 1996.

.... the interaction of Web servers and operating systems [Almeida et al. 1996; Yates et al. 1997; Schechte and Sutaria 1996] Kaashoek et al. have developed a prototype high performance Web server based on a customized operating system that is tailored specially for servers [Kaashoek et al. 1997; Kaashoek et al. 1996]. The request generation method used in all these studies is similar to the simple method of Section 3. There is also much existing work towards characterizing the invariants in WWW traffic. Arlitt and Williamson [Arlitt and Williamson 1996] characterized several aspects of Web server workloads ....

Kaashoek, M. F., D. R. Engler, G. R. Ganger, and D. A. Wallach (1996), "Server Operating Systems," In Proceedings of the 1996 ACM SIGOPS European Workshop, Connemara, Ireland, pp. 141--148.

....the file modification time, file length, etc. We are currently trying to pre compute the HTTP head information and store the information in the file cache when a file is loaded (or reloaded) into the file cache. This should eliminate most overhead caused by computing the HTTP response header. In [28] Kaashoek et al. also suggested to pre compute the HTTP header information 14 and store it in a file. Our scheme does not have to store the headers in files. Rather, it takes the advantage of our user level cache and automatically calculates the header information when a file is loaded into the ....

....of the Web requests. The scheme is effective and simple to implement. The idea of directly transferring data between the disk system and the network system has been proposed before in the Scout OS [32] in the Container Shipping system [33] in the IO Lite system [34] and in the MIT Server OSs [28]. Our profiling results strongly support applying the idea to Web servers, because Web servers spend most of their CPU time on copying data. 8 Conclusions In this paper, we present measurement results and performance analyses of the behavior of the Apache Web server on a uniprocessor system and ....

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach, "Server operating systems," in 1996 SIGOPS European Workshop, Sept. 1996.

....true concurrency in the stream of tasks. Thus, on multiprocessor systems, event driven servers have as many threads as processors. Examples of event driven servers include Squid[CDN 96, Squ] and its commercial version NetCache[Net] Zeus[Zeu] thttpd[tht] and several research servers[BDR97, KEGW96, PDZ97] 2.2 select( The select system call allows a user process to wait for events on a set of descriptors. A process can indicate interest in three types of events on a descriptor: events that make a descriptor readable, those that make it writable, and exception events. This information is ....

M. Frans Kaashoek, Dawson R. Engler, Gregory R. Ganger, and Deborah A. Wallach. Server Operating Systems. In Proceedings of the 1996 SIGOPS European Workshop, pages 141--148, Connemara, Ireland, September 1996.

....may need only optimized access to the raw device to store structured data. These applications may use access methods, similar to the IBM mainframe ISAM and VSAM access methods [IBM93] Another approach for achieving high performance on commodity hardware is through server operating systems [Kaashoek96], which provide abstractions and runtime support for specialized, high performance server applications. This approach of having specialized operating systems for servers is close to our approach of having specialized file systems for appliances. 2 Our position There are many capabilities lacking ....

M. Frans Kaashoek, Dawson R. Engler, Gregory R. Ganger, and Deborah A. Wallach. Server operating systems. Proceedings of the 1996 SIGOPS European Workshop (Connemara, Ireland), pages 141-- 148, September 1996.

.... web server performance [11] Likewise, there has been considerable prior work in performance analysis and benchmarking of conventional web servers [15, 12, 17, 5, 18] performance optimizations of web servers, performance oriented web server design, and operating system support for web servers [13, 22, 6, 7, 21]. Apostolopuolos et al. 3] studied the cost of TLS connection setup, RC4 and MD5, and proposed TLS connection setup protocol changes. Our methodology is to replace each individual operation within TLS with a no op and measure the incremental improvement in server throughput. This methodology ....

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach. Server operating systems. In Proceedings of the

....a multiprocessor implementation of protected sharing of TCP sockets among untrusted processes is viable. 45 4.4. 1 Organization of An User level TCP Socket Implementation Our TCP sockets are based on an existing library implementation of TCP, used in earlier exokernel networking applications [24, 23]. The programming interface, similar to that of traditional UNIX, offers eight procedures of significance: socket, bind, listen, connect, accept, read, write, and close. Their semantics are described below. The socket procedure creates a data structure, called tcpsocket info, that contains a ....

M. F. Kaashoek, D. Engler, G. Ganger, and D. Wallach. Server operating systems. In Proceedings of the 1996 SIGOPS European Workshop, pages 141--148, September 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141-148, 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop, September 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141--148, 1996.

No context found.

M. Kaashoek, D. Engler, D. Wallach, and G. Ganger. Server operating systems. In SIGOPS European Workshop '96, pages 141-148, 1996.

No context found.

M. F. Kaashoek, D. R. Engler, G. R. Ganger, and D. A. Wallach. Server operating systems. In Proceedings of the 1996.

No context found.

M. F. Kaashoek, D. Engler, G. Ganger and D. Wallach. "Server Operating Systems", in Proc. SIGOPS European Workshop, ACM, Sept. 1996.

First 50 documents

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