| Vivek S. Pai, Peter Druschel, and Willy Zwaenepoel. I/O-Lite: A Unified I/O Buffering and Caching System. In Proceedings of the Third USENIX Symposium on Operating Systems Design and Implementation, pages 15--28, February 1999. |
....poll( have been improved by [13, 14] to reduce event notification overhead. Another approach is to define completely new interfaces, building Web servers around these new interfaces. For example, new user mode interfaces to eliminate memory copies and mitigate checksum computation include IOLite [15] and Windows NT s TransmitFile( API. IO Lite provides a generic interface and mechanism to unify data management among operating system subsystems and user mode servers. TransmitFile( provides the same performance effect in avoiding data copies, but is limited to sending files with prefix or ....
Vivek S. Pai, Peter Druschel, and Willy Zwaenepoel. IOlite: A unified I/O buffering and caching system. In Operating Systems Design and Implementation (OSDI '99), pages 15--28, 1999.
....the longevity of bindings between host and payload cache buffers by forwarding data without potential copies into user space. Copy reduction techniques similarly impact the lifetime of bindings. These include page remapping, RDMA, and I O buffering support for scatter gather such as IO Lite [13]. For example, because IO Lite buffers exist as singlets and are immutable, writes create new host buffers and do not invalidate payload cache bindings. Most invalidations then occur when payload cache entries are replaced. A few interesting applications and future research directions of payload ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. IOLite: A unified I/O buffering and caching system. In Proceedings of the Third Symposium on Operating Systems Design and Implementation (OSDI'99), February 1999. 12
....technique relies on shared memory semantics between the user and kernel address space and permits to use DMAs for moving data between the shared memory and network interface. Such drivers can also be built with per process buffer pools that are pre mapped in both the user and kernel address spaces [7, 17]. 3. User Kernel Page Remapping with Copy on Write: These implementations re map memory pages between user and kernel space by editing the MMU table and perform copies only when 2 needed. They can also benefit from DMA to transfer frames between kernel buffers and the network interface [5] ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. I/O-Lite: A Unified I/O Buffering and Caching System. In Proceedings of the Third Symposium on Operating Systems Design and Implementation (OSDI '99), pages 15--28, 1999.
....virtual interfaces. 3. User Kernel Shared Memory: This scheme defines a new set of APIs with shared semantics between the user and kernel address space and uses DMA to move data between the shared memory and network interface. Proposals in this category are fast buffers (fbufs) 9] and IO Lite [20]. It uses per process buffer pools that are pre mapped in both the user and kernel address spaces, thus eliminating the user kernel data copy. 4. User Kernel Page Remapping with Copy on Write: This approach re maps memory pages between user and kernel space. At the sender the user data is mapped ....
Vivek S. Pai, Peter Druschel, and Willy Zwaenepoel. I/O-Lite: A Unified I/O Buffering and Caching System. In Proceedings of the Third Symposium on Operating Systems Design and Implementation (OSDI '99), pages 15--28, 1999.
....interface that supports checksum offloading. Operating systems that support zero copy I O use special APIs or memory management schemes to avoid copying data between the kernel and user space of main memory or between different subsystems of the kernel (such as the file cache and network buffers) [14, 26, 29]. As described in Section 2.2, network interfaces that support checksum offloading reduce load on the host CPU by directly calculating IP header checksums and TCP packet checksums so that the operating system running on the host CPU need not compute checksums. Both techniques aim to reduce load on ....
....In particular, they emphasize importance of zero copy I O support in web servers. They report that a zero copy sendfile system call implemented in IBM AIX workstations increased web server throughput by up to 51 . Pai et al. developed a unified I O buffering and caching system called IO Lite [29]. Like fbufs, IO Lite is based on immutable buffers and provides zero copy I O across protection domains. Unlike fbufs, IO Lite provides buffering and caching across applications and all subsystems of operating system including network, filesystem, file cache, and interprocess communication 62 ....
Vivek S. Pai, Peter Druschel, and Willy Zwaenepoel. I/O-Lite: A Unified I/O Buffering and Caching System. In Proceedings of the Third USENIX Symposium on Operating Systems Design and Implementation, pages 15--28, February 1999.
....system uses a PCI bus, network interface data caching does not depend on the specifics of the local interconnect. Network interface data caching exploits techniques originally developed for zero copy I O schemes, which avoid duplicating data between the kernel and user space of main memory [7, 11, 13]. Caching data at the network interface essentially extends the benefits of zero copy I O to the final crossing in the server between the memory and the network interface card (NIC) This technique alleviates the local interconnect bottleneck, allowing application level performance to scale with ....
....interface that supports checksum offloading. Operating systems that support zero copy I O use special APIs or memory management schemes to avoid copying data between the kernel and user space of main memory or between different subsystems of the kernel (such as the file cache and network buffers) [7, 11, 13]. Network interfaces that support checksum offloading reduce load on the host CPU by directly calculating IP header checksums and TCP packet checksums, allowing the operating system to transfer packets to and from the NIC without computing any checksums [9] Figure 1 shows the steps taken by a ....
[Article contains additional citation context not shown here]
Vivek S. Pai, Peter Druschel, and Willy Zwaenepoel. I/O-Lite: A Unified I/O Buffering and Caching System. In Proceedings of the Third USENIX Symposium on Operating Systems Design and Implementation, pages 15--28, February 1999.
.... 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 ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. I/O-Lite: A unified I/O buffering and caching system. In Proc. 3rd USENIX Symp. on Operating Systems Design and Implementation, New Orleans, Louisiana, Feb. 1999.
No context found.
PAI,V.S.,DRUSCHEL, P., AND ZWAENEPOEL, W. 1999b. I/O-Lite: A unified I/O buffering and caching system. In Proceedings of the 3rd USENIX Symposium on Operating Systems Design and Implementation (OSDI 99, New Orleans, LA., Feb.), USENIX Assoc., Berkeley, CA, 15--28.
....of Web server workloads [11, 23] Data copying within the operating system is a significant cost when processing large files, and several approaches have been proposed to alleviate the problem. Thadani et al. 30] introduce a new API to read and send memory mapped files without copying. IOLite [29] extends the fbufs [14] model to integrate filesystem, networking, interprocess communication, and application level buffers using a set of uniform interfaces. Engler et al. 20] use low level interaction between the Cheetah Web server and their exokernel to eliminate copying and streamline ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. IOLite: A unified I/O buffering and caching system. In Proceedings of the 3rd Symposium on Operating Systems Design and Implementation, New Orleans, LA, Feb. 1999.
....of Web servers, since these must often scale to thousands or millions of users. Operating systems researchers and system vendors have devoted much attention to improving the performance of Web servers. Improvements in operating system performance have come from reducing data movement costs [2, 35, 43], developing better kernel algorithms for protocol control block (PCB) lookup [26] and file descriptor allocation [6] improving stability under This paper will appear in the Proceedings of the 3rd USENIX Symposium on Operating Systems Design and Implementation (ODSI) New Orleans, LA, Feb. ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. IOLite: A unified I/O buffering and caching system. In Proc. 3rd Symp. on Operating Systems Design and Implementation, Feb. 1999.
....Pai et al. 23] The costs for the basic request processing steps used in our simulations were derived by performing measurements on a 300 MHz Pentium II machine running FreeBSD 2.2.6 and either the widely used Apache 1.3. 3 Web server, or an aggressively optimized research Web server called Flash [24, 25]. Connection establishment and teardown costs are set at 278 129 s of CPU time each, per request overheads at 527 159 s, and transmit processing incurs 24 24 s per 512 bytes to simulate Apache Flash, respectively. Using these numbers, an 8 KByte document can be served from the main memory cache ....
V. S. Pai, P. Druschel, and W. Zwaenepoel. I/O-Lite: A unified I/O buffering and caching system. In Proceedings of the 3rd Symposium on Operating Systems Design and Implementation, New Orleans, LA, Feb. 1999.
No context found.
Pai V., Druschel P., and Zwaenopoel, W., IOlite: A Unified I/O Buffering and Caching System, TR97-294, Rice University, 1997.
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