Nearly all IPUMS International samples, and all India samples, consider households to be the primary sampling unit (PSU). Therefore, it is usually advised to use the household identifier variable SERIAL for clustering. Note, however, that depending on the type of analysis you are performing, it may be more appropriate to cluster a different levels. For example, if you are aggregating data across regions before performing your analysis, then controlling for inter-region correlation of your outcome variable may be advisable. More details about variance estimation with IPUMS International data are available here. Specific sample characteristics for Indian samples are available here.