I do have some random parameters.i just curious what will lead to this
If you have been generating different random input data instantiations for different program runs, then of course you have been solving different optimization problems, even if of the same form; and therefore you can not expect to necessarily get the same results. See for instance http://ask.cvxr.com/t/how-to-sovle-this-qp-problem-in-cvx/695/3
In any event, CVX';s successive approximation method using SDPT3 is not doing well on the problem(s) you have provided, either failing completely or ending with Inaccurate/Solved (look at the final Status). You can try sedumi to see whether that works better, or MOSEK if you have that capability. Alternatively, you can try CVX 3.0beta with scs or ecos which will avoid the successive approximation method,and may do better.