P. Druschel. Operating systems support for highspeed networking. University of Arizona Ph.D. Dissertation CS-94-24, August 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....network interface into main memory, are copied from DMA buffers to stream buffers, 1 and then from stream buffers to user buffers. The problem with copying is that it incurs a per byte overhead. As network bandwidth increases, it is approaching the same order of magnitude as memory bandwidth [5]. Therefore, memory bandwidth may become the communication bottleneck if memory copies are not carefully minimized. Data flows generated by multimedia applications are more predictable than conventional data traffic. In a server for video playback, for example, a picture is made available for ....

....been proposed in [3] where requests to send are polled in the clock interrupt handler. Issues such as order of interleaving of sends by different user processes as well as impact on end to end delay are not considered in [3] An approach similar to direct media streaming has been considered in [5]. We have implemented an audio write path in SunOS 5.3 and studied its performance. Furthermore, we have described and evaluated two specific applications of our architecture for supporting distributed multimedia. ....

Peter Druschel. Operating system support for high-speed networking. Technical Report 94-24 (Ph.D. Dissertation), The University of Arizona, Tucson, Arizona, August 1994.

....more readily identified, and easier to resolve. Consequently, meeting the quality of service expectations of users is becoming a considerable challenge for today s administrators. In recent years, significant strides have been made towards the specification and provisioning of quality of service [4, 5, 6, 8, 9, 10, 11, 12, 15, 17, 18, 19, 26, 27, 31, 34, 37, 40, 44, 45, 46, 49, 50, 51]. As noted in [47] The human user of a multimedia application is the starting point for overall QoS considerations . Much of the work in this area recognizes this, and to varying degrees of success, aim to provide quality of service to meet user expectations. However, none of these approaches to ....

P. Druschel. Operating System Support for High-Speed Networking. PhD thesis, The University of Arizona, August 1994.

....in host memory to which the reassembled PDU should be transferred. The second interrupt is generated after the PDU has been copied to that buffer. Disadvantages of these adapters include the need for expensive buffers onboard to store PDUs and the time consumed in servicing interrupts. Druschel[13] describes strategies to reduce the number of interrupts generated to less than one per PDU. Just in time DMA adapters do not buffer PDUs in adapter memory. They are segmented from or reassembled into host memory directly. However, they are copied 5 twice as do buffering adapters. The number of ....

....a 50 100 improvement in processor performance annually since 1985 as compared to only a 7 improvement in memory performance. This disparity can lead to memory bottlenecks while transferring data[12] Transfer of data requires a memory to memory as well as a memory to adapter copy. Druschel[13] states that software data copying as a means of transferring data across domain boundaries exacerbates the memory bottleneck problem . The 6 domains refer to user and kernel address spaces. Druschel explains that these crossdomain transfers, which constitute the memory to memory copy across ....

[Article contains additional citation context not shown here]

Druschel, P., Operating System Support for High-Speed Networking, Ph.D. Dissertation, 1994. ftp://ftp.cs.arizona.edu/reports/1994/TR94-24.ps

....becoming a considerable challenge for today s administrators. It has become evident in recent years that the current approaches to resource management taken in systems software are not suitable to handle this kind of stress. The literature in this area provides considerable testimony to this fact [5, 9, 12, 24, 33, 38, 49]. Consequently, there has been considerable research effort towards developing better approaches to resource allocation and scheduling, with several promising results already realized. Nevertheless, many difficult outstanding problems remain to be addressed. One particular problem that has proven ....

....developing better approaches to resource allocation and scheduling to support quality of service. In this section, we take a look at some of this work. There has been significant work done towards improving resource scheduling and allocation of system and network resources for quality of service [4, 5, 6, 7, 8, 9, 10, 12, 17, 21, 22, 23, 24, 29, 30, 32, 34, 37, 38, 42, 43, 45, 47, 49, 50]. Most, if not all, of the related work in this area require applications or their users to specify how to achieve quality of service expectations through low level resource statements, however [5] Specifications of quality of service are typically done through a statement of the execution rate, ....

P. Druschel. Operating System Support for High-Speed Networking. PhD thesis, The University of Arizona, August 1994.

....the network interface into main memory, are copied from DMA buffers to stream buffers, 1 and then from stream buffers to user buffers. The problem with copying is that it incurs a per byte overhead. As network bandwidth increases, it is approaching the same order of magnitude as memory bandwidth [5]. Therefore, memory bandwidth may become the communication bottleneck if memory copies are not carefully minimized. Data flows generated by multimedia applications are more predictable than conventional data traffic. In a server for video playback, for example, a picture is made available for ....

....been proposed in [3] where requests to send are polled in the clock interrupt handler. Issues such as order of interleaving of sends by different user processes as well as impact on end to end delay are not considered in [3] An approach similar to direct media streaming has been considered in [5]. We have implemented an audio write path in SunOS 5.3 and studied its performance. Furthermore, we have described and evaluated two specific applications of our architecture for supporting distributed multimedia. ....

Peter Druschel. Operating system support for high-speed networking. Technical Report 94-24 (Ph.D. Dissertation), The University of Arizona, Tucson, Arizona, August 1994.

.... on aspects of interface design that allow messages to move more directly between application memory and message buffers in the network interface [4] 5] 6] 7] and on operating system and memory management issues that allow messages to cross protection domains without requiring a data copy [5] 8][9][10] 11] However, an actual zero pass system does not exist for transport layer Internet messages. A significant source of message latency remains: an embedded transport layer checksum induces a perbyte message latency that is not addressed by previous techniques. For Internet protocols, to date, ....

....input output (PIO) or by DMA. PIO is more flexible, and in some systems PIO is effectively faster than DMA. However, the reason often mentioned for the effectiveness of PIO is that it allows checksum calculation to be combined with data copying at no additional cost in RISC architectures [9]; this is no longer an advantage with a zero pass checksum. Without this advantage, there is little reason to commit processor cycles to transfer internet messages. The use of DMA to move data to from network channels will allow the CPU to perform other tasks. However, DMA engines access system ....

[Article contains additional citation context not shown here]

Druschel, P. Operating System Support for High-Speed Networking TR 94-24, Dept. of Computer Science, University of Arizona, August 1994.

....75 77 224 99 73 72 46 276 640 114 276 640 Figure 1 Memory and I O speeds for the DEC 3000 model 800. Sustained copy rates include TLB misses and imperfect cache behavior, plus interrupt overhead from a running, though otherwise idle, Unix kernel. Burst transfer rates are higher [Dut 92] [Dru94]. 3 Container Shipping is a fast I O system suitable for use with any operating system. Container Shipping includes a user level API (Application Programming Interface) for copy free I O. Like the related systems discussed below, our transfer mechanism is based on VM (Virtual Memory) remapping, ....

Peter Druschel, "Operating System Support for High-speed Networking", Ph.D. Dissertation, University of Arizona, August 1994.

....Genie also has several techniques for avoiding data fragmentation when the input data contains packet headers from network interfaces. The Fbuf IPC Facility The fast buffers (fbufs) kernel subsystem is an operating system facility for IPC buffer management developed at University of Arizona [23, 24, 65]. Fbufs provide fast data transfer across protection domain boundaries. The fbuf system is based on the assumption that IPC buffers are immutable. An immutable buffer is one that is not modified after it has been initialized. An example of an immutable buffer is a buffer that has been queued for ....

P. Druschel. Operating System Support for High-Speed Networking. PhD thesis, University of Arizona, August 1994.

....or a hashing scheme can be used to minimize this overhead, but in any case, it is not amenable to easy hardware implementation. We would like to point out here that the details of the types of checks performed, or the mechanism used, has not been elaborated upon in ADC or U Net publications [9,11,14]. Our approach to solving the problem of providing untrusted user space protocols the ability to queue or dequeue data directly to from the interface is called Protected DMA. This approach does not suffer from any of the three drawbacks mentioned above, and it is usable by both APIC and ....

Druschel, P., "Operating System Support for High-Speed Networking," University of Arizona Ph.D. Dissertation CS-94-24, Aug. 1994.

....for broadcast TV quality video, would provide about 1 Mbyte of data per second. And as network bandwidth increases, it is approaching the same order of magnitude as memory bandwidth. Therefore, memory bandwidth may become the communication bottleneck if memory copies are not carefully minimized [5]. Second, data flows generated by multimedia applications have characteristics that are more predictable than conventional data traffic. In a server for video playback, for example, pictures are made available for sending once every frame period. To transfer control between kernel and user for ....

....[3] where requests to send are polled in the clock interrupt handler. Some issues considered herein, such as order of interleaving of sends by different user processes as well as impact on end to end delay, were not considered in [3] An approach similar to direct media streaming was mentioned in [5]. We have investigated the idea more carefully. In particular, we have implemented an audio write path in SunOS 5.3 and studied its performance. Lastly, we have described and evaluated two particular applications of our architecture for supporting distributed multimedia. ....

Peter Druschel. Operating system support for high-speed networking. Technical Report 94-24 (Ph.D. Dissertation), The University of Arizona, Tucson, Arizona, August 1994.

....Our LRP architecture represents a more fundamental departure from the interrupt driven model, and rules out receiver livelock by design. In addition, LRP eliminates traffic interference and improves fairness and throughput under high load. The U Net system [1] and Application Device Channels (ADC) [3, 2] share with our LRP architecture the approach of using the network interface to demultiplex incoming packets and place them on queues associated with communication endpoints. With U Net and ADCs, the endpoint queues are mapped into the address space of application processes. Network processing is ....

P. Druschel. Operating systems support for high-speed networking. Technical Report TR 94-24, Department of Computer Science, University of Arizona, Oct. 1994.

....protection boundaries from the critical send receive path, and or by enabling application specific customization of protocol services. To the best of our knowledge, the behavior of user level network subsystems under overload has not been studied. U Net [1] and Application Device Channels (ADC) [4, 5] share with NI LRP the approach of using the network interface to demultiplex incoming packets and placing them on queues associated with communication endpoints. With U Net and ADCs, the endpoint queues are mapped into the address space of application processes. More conventional user level ....

P. Druschel. Operating systems support for highspeed networking. Technical Report TR 94-24, Department of Computer Science, University of Arizona, Oct. 1994.

No context found.

P. Druschel. Operating systems support for highspeed networking. University of Arizona Ph.D. Dissertation CS-94-24, August 1994.

No context found.

P. Druschel. Operating systems support for highspeed networking. University of Arizona Ph.D. Dissertation CS-94-24, August 1994.

No context found.

Druschel, P., "Operating System Support for High-Speed Networking," University of Arizona Ph.D. Dissertation CS-94-24, Aug. 1994.

No context found.

[Dru94a]Druschel P., Operating system support for high-speed networking. Ph.D. Dissertation, Department of Computer Science, University of Arizona (August 1994).

No context found.

Peter Druschel. Operating system support for high-speed networking. Technical Report 94-24, Department of Computer Science, University of Arizona, Tucson, AZ, USA, August 1994. Ph.D. Dissertation.

No context found.

Peter Druschel. Operating system support for high-speed networking. Technical Report 94-24 (Ph.D. Dissertation), The University of Arizona, Tucson, Arizona, August 1994.

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