... | ... | @@ -128,17 +128,18 @@ a separate view service) are uncommon in practice. |
|
|
|
|
|
The primary/backup scheme in this lab is not based on any published protocol. In
|
|
|
fact, this lab doesn't specify a complete protocol; you must flesh out the
|
|
|
protocol. The protocol has similarities with [Flat Datacenter
|
|
|
Storage][flat-storage] (the viewservice is like FDS's metadata server, and the
|
|
|
primary/backup servers are like FDS's tractservers), though FDS pays far more
|
|
|
attention to performance. It's also a bit like a MongoDB replica set (though
|
|
|
MongoDB selects the leader with a Paxos-like election). For a detailed
|
|
|
description of a (different) primary-backup-like protocol, see [Chain
|
|
|
Replication][chain]. Chain Replication has higher performance than this lab's
|
|
|
design, though it assumes that the view service never declares a server dead
|
|
|
when it is merely partitioned. See [Harp][harp] and [Viewstamped
|
|
|
Replication][vr] for a detailed treatment of high-performance primary/backup and
|
|
|
reconstruction of system state after various kinds of failures.
|
|
|
protocol. The protocol has similarities with
|
|
|
[Flat Datacenter Storage][flat-storage] (the viewservice is like FDS's metadata
|
|
|
server, and the primary/backup servers are like FDS's tractservers), though FDS
|
|
|
pays far more attention to performance. It's also a bit like a MongoDB replica
|
|
|
set (though MongoDB selects the leader with a Paxos-like election). For a
|
|
|
detailed description of a (different) primary-backup-like protocol, see
|
|
|
[Chain Replication][chain]. Chain Replication has higher performance than this
|
|
|
lab's design, though it assumes that the view service never declares a server
|
|
|
dead when it is merely partitioned. See [Harp][harp] and
|
|
|
[Viewstamped Replication][vr] for a detailed treatment of high-performance
|
|
|
primary/backup and reconstruction of system state after various kinds of
|
|
|
failures.
|
|
|
|
|
|
|
|
|
### Collaboration Policy
|
... | ... | |