As Jey has so eruditely pointed out, this advice borders on useless. I've always implemented LL queues as a segment of a list, where each element is protected by it's own mutex wherein each thread may add or remove a node by taking a lock on the first node, then moving to the second node, letting go of the first lock and so on. Although you'll end up holding more than one lock while traversing and three while deleting, this is at the core of powerful ideas like hand-over-hand locking.