"Glacier: Highly durable, decentralized storage despite massive correlated failures"

REVIEW: "Glacier: Highly durable, decentralized storage despite massive
correlated failures"

The paper presents Glacier a decentralized distributed storage system.
Glacier has been designed to mask large-scale correlated failures caused,
for example by network failures or malicious attacks, and this is achieved
by means of massive replication. To reduce storage requirements as well as
network overhead, Glacier uses erasure coding. It also employs garbage
collection to further reduce storage requirements. Glacier assumes that
the lifetime of a node is sufficiently large enough to justify replicating
data at the node. This peer-to-peer system uses consistent hashing wherby
keys are assigned to online nodes or nodes that were online within a given
time period.

Experimentaly, the authors found that their system can provide six nines
even when 60% of nodes experience correlated failures, at the cost of
eleven times the storage overhead. The system can be configured to prevent
data loss even when 85% of nodes are unavailable.

Although some consideration has been made to reduce the impact of
replication, this system imposes a high overhead in the network thus
providing limited scalability. From experimental results, it is shown that
nodes can experience orders of 500MB data per day.
Received on Thu Nov 24 2005 - 10:17:34 EST