The paper studies the input queued switch that is a good model for data center networks, operating under the MaxWeight algorithm. The heavy-traffic scaled mean queue length was characterized in Maguluri et al. (2018), Maguluri and Srikant (2016) when the arrivals are i.i.d.. This paper characterizes the heavy-traffic scaled mean sum queue length under Markov modulated arrivals in the heavy-traffic limit, and shows that it is within a factor of less than 2 from a universal lower bound. Moreover, the paper obtains lower and upper bounds, that are applicable in all traffic regimes, and they become tight in heavy-traffic regime.The paper obtains these results by generalizing the drift method that was developed in Eryilmaz and Srikant (2012), Maguluri and Srikant (2016) to the case of Markovian arrivals. The paper illustrates this generalization by first obtaining the heavy-traffic mean queue length in a single server queue under Markovian arrivals. The paper also illustrates the generalization of the transform method Hurtado-Lange and Maguluri (2018) to obtain the queue length distribution in a single server queue under Markovian arrivals. The key ideas are the use of geometric mixing of finite-state Markov chains, and studying the drift of a test function over a time window that depends on the heavy-traffic parameter.