P. B. Danzig and S. Melvin, `High resolution timing with low resolution clocks and a microsecond resolution timer for Sun workstations', Operating Systems Review, 24(1), 23--26 (1990).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of accuracy, NTP must be tuned correctly and allowed to stabilize for one to two weeks. 2.1.2.5 Using low resolution clocks During the 1980 s, the resolution of system clocks was on the order of 10 milliseconds. Peter Danzig addressed the problem of using a clock with insufficient resolution in [DM88]. He demonstrated how to use repetition and statistical methods to report results three significant digits beyond the clock resolution. Later systems generally provide microsecond resolution, but Danzig s technique is still useful when the accuracy of the clock is in the millisecond range. For ....

....Using stat ist ics If it is not possible to stabilize the clock, accurate data collection is still possible using statistical methods. We will describe two approaches. In the first approach, the resolution of the timestamps are reduced to the level of accuracy of the clock. Then Peter Danzig s [DM88] technique is applied for collecting data at accuracy levels beyond the clock resolution. The second approach retains the information in the high resolution timestamps. Each timestamp is treated as an estimate of the actual time and the Central Limit Theorem is used to assess the accuracy of the ....

[Article contains additional citation context not shown here]

Danzig, P.B. and S. Melvin. High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations. ACM Operating Systems Review 24, 1 (Jan) 1988, 23-26.

....Internet into the tens of msec range [Mil90, Mil94] However, to do this, NTP must be allowed to run undisturbed for one to two weeks. During that time, NTP itself is a major source of clock instability. The standard solution for dealing with unstable system clocks is to build a hardware clock [Mog90, CDJM91, DM88]. Hardware clocks are inherently more stable and sidestep most of the jitter problems of software clocks. In [DM88] Peter Danzig demonstrated how to use statistical methods to report results at higher resolutions than the clock. In that paper, he addressed the issue of a stable clock with ....

....weeks. During that time, NTP itself is a major source of clock instability. The standard solution for dealing with unstable system clocks is to build a hardware clock [Mog90, CDJM91, DM88] Hardware clocks are inherently more stable and sidestep most of the jitter problems of software clocks. In [DM88], Peter Danzig demonstrated how to use statistical methods to report results at higher resolutions than the clock. In that paper, he addressed the issue of a stable clock with insufficient resolution, which was the case with most systems before LANs. Now we have the opposite problem, since most ....

Danzig, P.B., and Melvin, S. "High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations." ACM Operating Systems Review 24, 1 (January 1988), 23-26.

....IP fragments accounted for only 0.02 of all IP packets at Berkeley, 0.05 at USC, and 0.02 at Bellcore. Submitted to Computer Networks and ISDN Systems December 15, 7 4.3. Tracing Instruments The Berkeley traces were gathered using a Sun 3 50 workstation equipped with a microsecond timer [6]. The Ethernet driver in the SunOS kernel was modified to write trace records to a circular kernel buffer big enough to hold 128 full size Ethernet packets. The resulting time stamp resolution was 10 microseconds. A dedicated user level program inspected the buffer and wrote new records to tape. ....

P. B. Danzig and S. Melvin, High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations, ACM Operating Systems Review 24, 1 (January, Submitted to Computer Networks and ISDN Systems December 15, - 15 - 1990).

....in [7] That paper described a single pseudo device for measuring file system activity. The measurement system described here also includes a high resolution timer attached to the Sun 3 50 SCSI port. A timer making use of the data encryption chip socket on Sun 3 and Sun 4 computers is described in [8]. Profiling of a network protocol stack with timestamps provided by an interval timer is reported in [9] 3 Design And Implementation Figure 1, below, shows an overview of the system. The name of the pseudo controller is logdev. It resides in kernel space on the target Sun 3 50, and manages a ....

P.B. Danzig and S. Melvin. High Resolution Timing with Low Resolution Clocks and A Microsecond Resolution Timer for Sun Workstations. In Operating Systems Review, Vol. 24, No. 1, pages 23-26, Jan. 90.

....traces were gathered using a Sun 3 50 workstation configured with a LANCE Ethernet controller, a SCSI disk drive, and a SCSI 8 millimeter tape drive. It was also equipped with a microsecond resolution timer board designed at Berkeley and connected through the unused DES data encryption port [19]. The Ethernet driver in the SunOS kernel was modified to manage a circular buffer big enough to hold 3,000 trace records. Immediately after an interrupt was received signaling the arrival of an Ethernet packet, the system read the current value of the timer, truncated the Ethernet packet to 56 ....

P. B. Danzig and S. Melvin, High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations, ACM Operating Systems Review 24, 1 (January, 1990).

....bound or the frequency of agreement protocol, or both. The intuitive idea is that using EFF each clock learns whether it is advancing or lagging with respect to the average and compensates for it. A third problem is making accurate timing measurements with a coarse interval timer. A recent report [1] describes two ways to make accurate measurements on SUN workstations, which have a 20 millisecond resolution interval timer. The first method is to take many measurements and average them. The drawback is that we need a large number (thousands to hundreds of thousands) of data points to achieve ....

P. Danzig and S. Melvin. High resolution timing with low resolution clocks and a microsecond resolution timer for sun workstations. ACM SIGOPS Operating Systems Review, 24(1):23--26, January 1990.

....Labs traces were not saved on tape. Rather, packet counts were broken down by application type and data unit length, and written to disk; the raw trace data was discarded. 4.4. Tracing Instruments The Berkeley traces were gathered using a Sun 3 50 workstation equipped with a microsecond timer [6]. The Ethernet driver in the SunOS kernel was modified to write trace records to a circular kernel buffer big enough to hold 128 full size Ethernet packets. The resulting time stamp resolution was 10 microseconds. A dedicated user level program inspected the buffer and wrote new records to tape. ....

P. B. Danzig and S. Melvin, High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations, ACM Operating Systems Review 24, 1 (January, 1990). Submitted to IEEE INFOCOM 92 June 28, - 13 -

....just turn NTP off. The pros and cons of this strategy are discussed in Section 4.1. 2.3 Previous Work on Insufficient Clocks Though most papers make no mention of the clock used, there are some where the authors are clearly aware of this problem. The standard solution is to build a hardware clock [Mog90, CDJM91, DM88]. Hardware clocks are inherently more stable and sidestep most of the jitter problems of software clocks. Two sources of instability that remain are accessing it in a deterministic way and the normal wander of a room temperature oscillator [Mil90] Oscillator wander is in the range of 0.1 to 1 ....

Danzig, P.B., and Melvin, S. High resolution timing with low resolution clocks and a microsecond timer for sun workstations. ACM Operating Systems Review 24, 1 (January 1988), 2326.

.... 0.2 0.1 1.8 0.1 other 4.9 11.3 1.6 0.4 3.1 3.1 1.1 2.2 2.4 Table 2: Breakdown of unidirectional TCP traffic, by packets, by bytes, and by conversations. 2.3. Tracing Instrumentation and Packet Loss Rate The UCB data was collected with a Sun 3 workstation equipped with a microsecond timer [Danzig90]. The resulting time stamp resolution was 10 microseconds. The workstation ran a modified Unix kernel with a circular buffer big enough to hold 128 full size Ethernet packets. A dedicated user program transferred trace records from this buffer to tape. No packet losses due to buffer overflows were ....

Danzig, P.B. and Melvin, S., "High Resolution Timing with Low Resolution Clocks and a Microsecond Timer for Sun Workstations," ACM OS Review, 24:1, Jan '90, pp. 23-26.

No context found.

P. B. Danzig and S. Melvin, `High resolution timing with low resolution clocks and a microsecond resolution timer for Sun workstations', Operating Systems Review, 24(1), 23--26 (1990).

No context found.

Peter B. Danzig and Steve Melvin. High resolution timing with low resolution clocks and a microsecond timer for Sun workstations. Operating Systems Review, 24(1):23-26, January, 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