Leases: An Efficient Fault-Tolerant Mechanism for Distributed File
Cache Consistency
--- Gray and Cheriton, 1989.


- with large caches, the traffic required to maintain consistency can
  be the dominant factor in cache performance;

- leases are proposed as a consistency protocol that handles host and
  communication failures using physical clocks;

- from V system measurements, it is shown that short-term leases
  provide near optimal efficiency;

* a lease is a contract that gives its holder specified rights over
* property for a limited period of time.  In the context of caches, a
* lease grants control over writes to the covered datum during the
* term of the lease;

* the server must obtain the approval of the leaseholder before the
* datum may be written;

- when a datum is fetched from the server, the server also returns a
  lease;

- multiple uses of the cached file during the term of the lease are
  executed with no further communication with the server;

* when a client writes a datum, the server must defer the request
* until each leaseholder has granted approval or the term of its lease
* has expired;

* if some host holding a lease for a file is unreachable, writes have
* to wait until the lease expires;  this is one of the advantages of
* short leases;  they also minimize false write-sharing;

- false sharing occurs when a client writes to a file which is covered
  by a lease held by an inactive client;  in this case, leases add the
  overhead of executing a callback to the leaseholder;

- long-term leases are good for files that are accessed repeatedly and
  show little write-sharing;

- given:

	- R (rate of reads);
	- W (rate of writes);
	- S (number of caches);
	- alpha (lease benefit factor);

  we have:

	alpha = 2R/SW

- a sufficient long lease term will reduce server load whenever alpha
  is greater than one.  Large values for R and alpha imply better
  performance for short terms;

- most of the benefit of a non-zero lease term is gained by a term of
  just a few seconds;

- the analytic model agrees with the V data set;

- leases in the order of 10 seconds show low server load and low added
  delay due to consistency;

- larger propagation delay between clients and servers means that the
  impact of lease extensions and invalidations on response time is
  greater;

- they increase the avg RTT to 100 miliseconds and conclude that
  leases in the range 10-30 seconds still appear to be adequate;

- increased numbers of clients and servers have no effect unless it
  increases the level of write-sharing;

- leases depend on well-behaved and loosely-synchronized clocks;