Reliability of the Minsk-1 Electronic Digital Computer
Abstract Generated abstract
This paper evaluates the operational reliability of the Minsk-1 electronic digital computer using statistical records from technical-operation and machine-time logs. It defines and applies standard reliability and use criteria, including mean time between failures, mean repair time, probability of failure-free operation, preventive-maintenance effectiveness, utilization, and availability. For the observed period, the mean time between computation-mode failures is estimated at 19.58 hours, with a 0.9 confidence interval of 17.48 to 22.25 hours, and the mean time to locate and eliminate a failure is 2.04 hours. The analysis finds high preventive-maintenance effectiveness and availability above 90%, identifies electronic tubes as the dominant source of failures, and suggests preliminary tube conditioning to reduce early-life failures.
Full Text
RELIABILITY OF THE “MINSK-1” ELECTRONIC DIGITAL COMPUTER
V. I. KIZEV, M. N. STROGONOV
(Presented by the scientific seminar of the Computing Laboratory of TPI)
One of the basic technical characteristics of an electronic digital computer is its reliability in the course of operation. The principal quantitative reliability criteria are [1—3]:
- The mean operating time between two adjacent failures
\[ T_{\mathrm{sr}}=\frac{\sum_{i=1}^{N_{\mathrm{v}}} t_{\mathrm{v}i}}{N_{\mathrm{v}}}, \tag{1} \]
where \(t_{\mathrm{v}i}\) is the time of fault-free operation in computation mode from the \(i-1\)-st to the \(i\)-th failure;
\(N_{\mathrm{v}}\) is the number of failures that occurred during operation of the electronic digital computer in computation mode.
The quantity \(\lambda_{\mathrm{v}}=\frac{1}{T_{\mathrm{sr}}}\) is called the failure rate of the electronic digital computer in computation mode.
- The mean time for locating and eliminating one failure during operation of the machine in computation mode
\[ t_{\mathrm{puo}}=\frac{\sum_{i=1}^{N_{\mathrm{v}}} t_{\mathrm{puo}i}}{N_{\mathrm{v}}}, \tag{2} \]
where \(t_{\mathrm{puo}i}\) is the time spent on locating and eliminating the \(i\)-th failure.
- The probability of fault-free operation
\[ P(t)=e^{-\frac{t}{T_{\mathrm{sr}}}}, \tag{3} \]
where \(t\) is the specified duration of operation of the electronic digital computer in computation mode.
- The coefficient of preventive-maintenance effectiveness
\[ K_{\mathrm{ep}}=\frac{N_{\mathrm{prof}}}{N_{\mathrm{prof}}+N_{\mathrm{v}}}, \tag{4} \]
where \(N_{\mathrm{prof}}\) is the number of possible failures prevented during preventive maintenance.
- Coefficient of preventive maintenance
\[ K_{\mathrm{p}} = \frac{\displaystyle \sum_{j=1}^{m} t_{\mathrm{prof}\,j}} {\displaystyle \sum_{i=1}^{N_{\mathrm{p}}} t_i}, \tag{5} \]
where \(t_{\mathrm{prof}\,j}\) is the time spent on carrying out the \(j\)-th preventive maintenance.
- Utilization coefficient
\[ K_{\mathrm{u}}=\frac{t_{\mathrm{v}}}{t_{\mathrm{v}}+t_{\mathrm{p}}} =\frac{t_{\mathrm{v}}}{t_{\mathrm{cal}}}, \tag{6} \]
where \(t_{\mathrm{v}}\) is the time during which the electronic digital computer was in the switched-on state over a specified calendar period.
- Readiness coefficient
\[ K_{\mathrm{r}}=\frac{t_{\mathrm{v}}}{t_{\mathrm{v}}+t_{\mathrm{puo}}} =\frac{\overline{T}_{\mathrm{sr}}}{\overline{T}_{\mathrm{sr}}+\overline{T}_{\mathrm{puo}}}. \tag{7} \]
Statistical data on the operation of the machine, obtained on the basis of entries in the technical-operation log of the “Minsk-1” electronic digital computer and in the machine-time accounting log, are brought together in Table 1.
From the data of Table [1], on the basis of formulas (1–7), criteria have been calculated that make it possible to judge the operational reliability and the effectiveness of use of the electronic digital computer (Table 2).
It should be noted that, when calculating the preventive-maintenance coefficient and the utilization coefficient by months, the time spent on carrying out semiannual and annual preventive maintenance is not taken into account. This time is taken into account when calculating the overall numerical criteria for the entire period under consideration.
On the basis of the statistical data for the entire period under consideration, the failures that occurred in the computation mode are distributed by devices as follows:
| UVV | UU | AU | MOZU |
|---|---|---|---|
| 45 | 29 | 14 | 12 |
Data for NML are not given because of the absence of a sufficient amount of statistical material.
The mean time between two adjacent failures during operation of the machine in computation mode, and the mean time for locating and eliminating a failure, with confidence intervals at confidence level \(\gamma=0.9\), have respectively the values:
\[ 17.48—22.25\ \text{hr.}, \]
\[ 1.82—2.32\ \text{hr.} \]
The confidence regions of possible values of the unknown parameters \(T_{\mathrm{sr}}\) and \(t_{\mathrm{puo}}\) are calculated from the inequalities [3]:
\[ T_{\mathrm{srmin}}= \frac{\overline{T}_{\mathrm{sr}}}{(+x_{\mathrm{sr}}\gamma)} \leq T \leq \frac{\overline{T}_{\mathrm{sr}}}{(-x_{\mathrm{sr}}\gamma)} =T_{\mathrm{srmax}} \tag{8} \]
\[ t_{\mathrm{puomin}}= \frac{\overline{t}_{\mathrm{puo}}}{(+x_{\mathrm{puo}}\gamma)} \leq t_{\mathrm{puo}} \leq \frac{\overline{t}_{\mathrm{puo}}}{(-x_{\mathrm{puo}}\gamma)} =t_{\mathrm{puomax}}. \tag{9} \]
According to the data, \(T_{\mathrm{srmin}}=17.48\) hr., \(T_{\mathrm{sr}}=19.58\) hr., \(T_{\mathrm{srmax}}=22.25\) hr., and using expression (3) we construct the graph (Fig. 1) of the region in which conclu-
Table 1
| Machine operation | Year, month | \(t_{\mathrm{prof}}\), hours | \(t_{\mathrm{to}}\), hours | \(t_{\mathrm{p}}\), hours | \(t_{\mathrm{z}}\), hours | \(t_{\mathrm{off}}\), hours | \(N_{\mathrm{v}}\) | \(N_{\mathrm{prof}}\) | \(N_{\mathrm{e}}\) | \(n_{\mathrm{p}}\) | Note |
|---|---|---|---|---|---|---|---|---|---|---|---|
| One-shift operation | 1963, IX | 70 | 7 | 77 | 70 | 147 | 3 | 19 | 22 | 23 | 48 hours — half-year preventive maintenance |
| 1 shift | X | 74 | 7 | 81 | 104,7 | 185,7 | 5 | 8 | 13 | 27 | 48 hours — half-year preventive maintenance |
| 1 shift | XI | 76 | 10,2 | 86,2 | 85,7 | 171,9 | 4 | 19 | 23 | 23 | 48 hours — half-year preventive maintenance |
| 1 shift | XII | 73,4 | 17 | 90,4 | 104,6 | 195 | 9 | 13 | 22 | 25 | 48 hours — half-year preventive maintenance |
| 1 shift | 1964, I | 73 | 7,2 | 80,2 | 80,5 | 160,7 | 3 | 6 | 9 | 20 | 48 hours — half-year preventive maintenance |
| 1 shift | II | 55 | 11 | 66 | 135,8 | 201,8 | 11 | 13 | 24 | 25 | 48 hours — annual preventive maintenance |
| 1 shift | III | 62 | 1,4 | 63,4 | 150,7 | 214,1 | 4 | 12 | 16 | 26 | 48 hours — annual preventive maintenance |
| 1 shift | IV | 36,3 | 7,6 | 43,9 | 141,3 | 185,2 | 7 | 11 | 18 | 25 | 48 hours — annual preventive maintenance |
| 1 shift | VIII | 88 | 52,4 | 140,4 | 91,4 | 231,8 | 4 | 13 | 17 | 20 | 48 hours — annual preventive maintenance |
| Total | 1964 | 607,7 | 120,8 | 728,5 | 964,7 | 1963,2 | 50 | 114 | 164 | 214 | All failures. Annual preventive maintenance |
| Two-shift operation | V | 61,7 | 6,5 | 68,2 | 256,5 | 324,7 | 9 | 13 | 22 | 23 | Annual preventive maintenance |
| Two-shift operation | VI | 92 | 14,8 | 106,8 | 217,3 | 324,1 | 11 | 15 | 26 | 26 | Annual preventive maintenance |
| Two-shift operation | VIII | 121,7 | — | 121,7 | — | 121,7 | — | — | — | — | Annual preventive maintenance |
| Two-shift operation | IX, X | 76,6 | 13,7 | 90,3 | 256,7 | 347,0 | 13 | 15 | 28 | 36 | 26 hours — half-year preventive maintenance |
| Two-shift operation | XI | 35 | 39,8 | 74,8 | 230,7 | 305,5 | 10 | 6 | 16 | 24 | 26 hours — half-year preventive maintenance |
| Two-shift operation | XII | 33,5 | 12,3 | 45,8 | 279 | 324,8 | 10 | 12 | 22 | 26 | 26 hours — half-year preventive maintenance |
| Two-shift operation | 1965, I | 60,2 | 16,8 | 77 | 230,6 | 307,6 | 9 | 30 | 39 | 18 | 26 hours — half-year preventive maintenance |
| Two-shift operation | II | 42,9 | 40,9 | 83,8 | 224,4 | 308,2 | 13 | 8 | 21 | 24 | 26 hours — half-year preventive maintenance |
| Two-shift operation | III | 138,1 | 41,4 | 179,5 | 245,4 | 424,9 | 30 | 25 | 55 | 25 | 26 hours — half-year preventive maintenance |
| Two-shift operation | IV | 67,1 | 19,4 | 86,5 | 278 | 364,5 | 11 | 18 | 29 | 24 | 26 hours — half-year preventive maintenance |
| Two-shift operation | V | 52,0 | 20,0 | 72,0 | 172,3 | 244,3 | 10 | 19 | 29 | 23 | 26 hours — half-year preventive maintenance |
| Two-shift operation | VI | 64,5 | 31,7 | 96,2 | 267,5 | 363,7 | 9 | 19 | 28 | 25 | 26 hours — half-year preventive maintenance |
| Total | 1965 | 845,3 | 257,3 | 1102,6 | 2658,4 | 3761 | 135 | 180 | 315 | 274 | 26 hours — half-year preventive maintenance |
| Total for the entire period | 1453 | 378,1 | 1831,1 | 3693,1 | 5454,2 | 185 | 294 | 479 | 488 |
Table 2
| Year, month | $\overline{T}_{\mathrm{avg}}$, h | $\overline{t}_{\mathrm{puo}}$, h | $K_{\mathrm{p}}$ | $K_{\mathrm{ep}}$ | $K_{\mathrm{g}}$ | $K_{\mathrm{u}}$ |
|---|---|---|---|---|---|---|
| 1963 IX | 23.33 | 2.33 | 1.0 | 0.86 | 0.909 | 0.476 |
| X | 20.94 | 1.4 | 0.707 | 0.61 | 0.936 | 0.564 |
| XI | 21.42 | 2.55 | 0.887 | 0.82 | 0.892 | 0.498 |
| XII | 11.7 | 1.88 | 0.702 | 0.68 | 0.860 | 0.536 |
| 1964 I | 26.83 | 2.4 | 0.309 | 0.66 | 0.918 | 0.50 |
| II | 12.34 | 0.91 | 0.405 | 0.52 | 0.925 | 0.673 |
| III | 37.67 | 0.35 | 0.411 | 0.750 | 0.991 | 0.704 |
| IV | 20.18 | 1.86 | 0.257 | 0.61 | 0.949 | 0.763 |
| VII | 22.85 | 13.1 | 0.438 | 0.71 | 0.636 | 0.476 |
| V | 28.5 | 0.72 | 0.240 | 0.591 | 0.975 | 0.79 |
| VI | 19.75 | 1.34 | 0.423 | 0.58 | 0.935 | 0.67 |
| — | — | — | — | — | — | — |
| IX, X | 19.75 | 1.05 | 0.298 | 0.536 | 0.940 | 0.730 |
| XI | 23.07 | 3.98 | 0.152 | 0.375 | 0.853 | 0.755 |
| XII | 27.9 | 1.23 | 0.12 | 0.545 | 0.958 | 0.859 |
| 1965 I | 25.6 | 1.87 | 0.148 | 0.759 | 0.932 | 0.819 |
| II | 17.3 | 3.15 | 0.191 | 0.381 | 0.846 | 0.728 |
| III | 8.18 | 1.38 | 0.563 | 0.454 | 0.856 | 0.578 |
| IV | 25.27 | 1.76 | 0.241 | 0.621 | 0.935 | 0.763 |
| V | 17.23 | 2.0 | 0.302 | 0.655 | 0.896 | 0.705 |
| VI | 29.72 | 3.52 | 0.241 | 0.679 | 0.894 | 0.736 |
| 1st shift | 19.29 | 2.42 | 0.755 | 0.695 | 0.8887 | 0.5697 |
| 2nd shift | 19.69 | 1.906 | 0.318 | 0.571 | 0.912 | 0.7068 |
| Total for the entire period | 19.58 | 2.04 | 0.401 | 0.614 | 0.906 | 0.664 |
a family of curves of the probability of failure-free operation of the electronic digital computer as a function of time was constructed.
The average time for locating and eliminating one failure during preventive maintenance is determined by the formula
\[ \overline{t}_{\mathrm{puo}} = \frac{t'_{\mathrm{prof.}} - t_{\mathrm{test.}}}{N_{\mathrm{prof.}}}, \]
where $t'_{\mathrm{prof.}}$ is the time spent on preventive maintenance of the machine, excluding annual and semiannual preventive maintenance;
\[ t_{\mathrm{test.}} = \overline{t}_{\mathrm{test.}} \cdot n_{\mathrm{prof.}} \]
is the time spent checking the machine with tests in the preventive-maintenance mode;
$\overline{t}_{\mathrm{test.}}$ is the average time spent on one check of the machine by a test in the “preventive-control” mode;
$n_{\mathrm{prof.}}$ is the number of preventive checks of the machine carried out over a specified calendar period.
The numerical values of the quantities under consideration are presented in Table 3.
Thus, the average time for finding and eliminating one failure during preventive maintenance of the machine is considerably less than during operation of the machine in the computation mode (Table 2), since, when the machine is operating in the “preventive check” mode, a failure is created artificially in the device being checked by disturbing the power-supply regime. It is much easier to localize than a failure that has arisen during operation of the machine in the computation mode.
Table 3
| $t'$ prev., h | $n$ prev. | $\bar t'$ test., h | $t$ test., h | $\bar t'_{\text{puo}}$, h |
|---|---|---|---|---|
| 1209.3 | 488 | 1.6 | 780.8 | 1.46 |
The average operating time of the machine between two adjacent failures after semiannual and annual preventive maintenances decreases somewhat.
Fig. 1. Probability of failure-free operation of the Minsk-1 electronic digital computer
This is explained by the fact that, during approximately 200–300 hours of machine operation after preventive maintenance, an intensive failure of tubes is observed.
Carrying out preventive maintenance after 7 hours (single-shift operation) or after 14 hours does not have a noticeable effect on the value of $T_{\text{av}}$. In both cases it is approximately the same. This is evidently explained by the fact that, although when preventive maintenance is carried out after 7 hours the number of possible gradual failures prevented is relatively greater than when preventive maintenance is carried out after 14 hours ($K_{\text{ep}}$ in the first case is greater than in the second), the total failure intensity $\lambda_{\text{v}}$ in both cases remains approximately the same owing to an increase in the intensity of sudden failures under single-shift operation of the machine.
As is noted in a number of works [4–8] and as operating experience shows, the main percentage of failures is accounted for by electronic tubes. Thus, according to statistics obtained on domestic computing machines MESM, BESM, “Strela,” “Ural,” “Kiev,” and foreign ones, tubes account for 50–65% of all observed damages.
Processing of the statistics for the Minsk-1 electronic digital computer over a period of 2643 hours of the machine’s switched-on state (November 1964–June 1965) showed
It turns out that electronic tubes account for about 83% of the failures in the machine’s replaced cells. Electronic tubes fail with great intensity during the initial period of their operation—the burn-in period. In order to eliminate the burn-in period, it is necessary to introduce preliminary conditioning of the tubes in the machine for 100–200 hours.
On the Minsk-1 electronic digital computer, tube conditioning was not carried out because a sufficient number of tubes was not available in reserve. The effectiveness of preventive maintenance is evaluated by the coefficient of effectiveness of preventive maintenance. For the Minsk-1 computer it is quite high. Faults detected during semiannual and annual preventive maintenance are not taken into account. The utilization coefficient, which most fully characterizes the effectiveness of use of an electronic digital computer, is approximately equal to the value of the utilization coefficient of tube machines of the Ural and Kiev types. The value of the availability coefficient, more than 90%, indicates a sufficiently high readiness of the computer in the computing mode.
References
-
M. V. Burakov. Experience in Operating the Ural Digital Computing Machine. Sov. Radio Publishing House, Moscow, 1962.
-
B. N. Malinovskii, S. S. Zabara. Studies of the reliability of the Dnipro general-purpose control machine. Avtomatika i priborostroenie, No. 4 (16), 1963.
-
N. A. Shishonok, V. F. Repkin, L. L. Barvinskii. Fundamentals of the Theory of Reliability and Operation of Radio-Electronic Equipment.
-
L. N. Dashevskii, S. B. Pogrebinskii, E. A. Shkabara. The Kiev Computing Machine. Tekhnika Publishing House, Kiev, 1964.
-
V. N. Semochkin. Organization of the Technical Operation of Radio-Relay Lines. Svyaz Publishing House, Moscow, 1965.
-
Ya. M. Ryabov. Increasing the reliability of operation of the Strela electronic computing machine. In: Reliability of Radio-Electronic Equipment. Sov. Radio Publishing House, Moscow, 1960.
-
A. L. Brudno, Yu. A. Lavrenok. Operation of the M-2 electronic digital computer. In: Digital Technology and Computing Devices. Publishing House of the Academy of Sciences of the USSR, Moscow, 1959.
-
V. A. Zimin. Reliability of tubes in an electronic computing machine. In: Ways of Development of Soviet Mathematical Machine Building and Instrument Making, Part II, Moscow, 1956.
-
L. P. Leont’ev, A. M. Morgulis. Reliability and service life of certain types of electronic tubes. Avtomatika i vychislitel’naya tekhnika, issue 6, Publishing House of the Academy of Sciences of the Latvian SSR, Riga, 1963.