Remote copying options

Last time, we discussed two aspects of data replication: volume copying and point-in-time (snapshot) copies.  Today let’s look at the third category of replication, remote copying.

Remote copies are particularly useful in disaster recovery scenarios.  Data is mirrored to a remote site over a campus, metropolitan-area network (MAN) or WAN, or in some cases over a dedicated line.  If the primary site goes down, failover software immediately targets the remote replicated volume, whose data is either identical or nearly identical to the primary volume.

The question of “identical” or “nearly identical” is an interesting one.  Which one do you want? That depends.

Nearly identical means you can live with the fact that writes to the replication volume somewhat lag writes to the primary system. Identical means that the two volumes are the same at any point in time, which requires synchronous data transfers.

Synchronous transfers means that a program writes to both volumes at once, but it also means that each data block being sent to the remote volume must be acknowledged to ensure the system’s integrity; failed transmissions must be corrected before a new one can be initiated.  To maintain a synchronous mirror, no data on the primary volume can be moved until the remote volume has verified that it has been updated.  Expect a performance hit there.

That’s problem No. 1 with synchronous.

Problem No. 2 has to do with the fact that no technology, however advanced, can break the laws of physics.  The speed of light doesn’t change within its transmission medium, so that electrons moving down a cable, be it copper or fiber optic, have a known maximum speed.  This means that distance implies latency – and the greater the distance, the greater the latencies involved. 

What follows from this is that writing data to a remote volume synchronously gets more challenging as the distance between the two volumes increases. For this reason, synchronous replication is generally limited to campus or MANs.

An interesting alternative approach to long-distance synchronous replication is the one taken by stealthy Canadian start-up Right Range Networks, which is researching a synchronous replication technique that calculates the distance to the replicated device, computes the latency, and slows down operations on the primary system to allow synchronous data transfers.  Prototypes of the system are being tested by the Nova Scotia provincial government’s Harp and Ribbon Seal Harvesting project.  Right Range CTO Duncan McKenzie says the company is engaged with several venture capital firms at this time.