When there's a large amount of discarded packets, it typically indicates either a runtime issue or some invalid queries in the data set. In both cases, the discarded packets can skew the results. If a serious runtime issue is encountered (e.g. max_clients exceeded), the result becomes misleading at best (or just worthless). Add a configurable limit for the ratio of discarded packets allowed in relation to total number of requests. In case this limit is exceeded, exit prematurely.