In recent years, a large number of DHTs have been proposed. However, very few of them have been deployed in real-life large scale systems. An exception is KAD, a DHT based on Kademlia that is part of the widely used eMule peer-to-peer system, which has more than 1.5 million simultaneous users. We have developed a very fast crawler and explored KAD for more than a month by crawling part of the KAD ID space, which takes only a few seconds. We find that there are two classes of peers, long-lived peers that participate in KAD for weeks and short-lived peers that remain in KAD no more than few days before they permanently leave. Most of the peers that join KAD for the first time are short-lived peers. Since inter-session times can be as large as one week, one needs to crawl KAD for more than a week to be able to identify and characterize short-lived peers.