It had been known since the late 18th century that the rotation of the earth was decelerating when measured with respect to the orbital motions of bodies in the solar system. Seasonal variations in the rotation of the earth were observed in the 1930s, and made it clear that Universal Time was inappropriate for the computation of planetary motions. In 1939 the IAU adopted a value of the Gaussian gravitational constant which defines the rate of ET, and the theoretical basis for ET was developed during the 1940s and 1950s and refined during the 1960s.
Cesium atomic clocks became operational in 1955, and quickly made it evident that the rotation of the earth fluctuated randomly. This confirmed the utter unsuitability of the mean solar second of Universal Time as a measure of time interval. After three years of comparisons with lunar observations it was determined that the ephemeris second corresponded to 9192631770 cycles of the cesium resonance. In 1960 the length of the SI second was defined to be equal to the ephemeris second.
Nevertheless, the theoretical basis for Ephemeris Time is wholly non-relativistic. In 1976 the IAU resolved that beginning in 1984 ET would be replaced by the two relativistic timescales Barycentric Dynamical Time (TDB) and Terrestrial Dynamical Time (TDT). For practical purposes the length of the ephemeris second can be taken as equal to the length of the TDB or TDT second.