This paper introduces a scheme of client device identification by clock skew in cloud environments. Clock skew of a remote device is considered an effective physical characteristic and is suitable for device identification purposes. As the temperature is always steady inside data center, we utilize this advantage to verify that two arithmetic relations hold with negligible measurement error between servers and develop a fault-tolerant multi-server skew measuring scheme. We have also implemented a lightweight application layer experimental platform to provide the device identification and account authentication service in an implicit manner based on the clock skew fingerprinting technique. Two experiments are conducted to confirm the correctness of skew arithmetic, and to examine the feasibility of client device identification by clock skew respectively. The experiment results show that the maximum error of additive inverse is only 0.01 ppm and the linearity of clock skew holds with error no more then 0.03 ppm while utilizing physical hardware. Furthermore, the capability of load balance and fault tolerance achieved by multiple time servers are also demonstrated in the experiments.