Justification :
This paper talks about Amazons new Key/Value storage system called
Dynamo. The author compares Dynamo with many of the present systems and
then shows how Dynamo is different from the rest .Primary theme of
paper was tradeoff between availability and consistency . Amazon has
chosen an always on eventually consistent model.The paper leverages
several techniques such as consistent hashing ,vector clock ,hinted
handoff,quorum systems at a brisk pace. Having been deployed and run at
a varied application environment at Amazon I feel Dynamo is highly
successful and meets essentially all the requirements and expectations
it was intended to meet.
Summery :
Dynamo is highly available highly scalable key value store.In order to
provide In order to provide high availability, the system adopts an
eventual consistency model and leverages several techniques like
consistent hashing for load balancing , preference list for data
replication, hinted handoff for availability , merkel tree for
synchronization and vector clock.Along with this the paper also provides
a performance analysis of real system application for various issues
like data replication and load balancing.
Pros :
a)The paper gives insight into how various different techniques can be combined to form a complex system.
b)Dynamo is a very practicable system.Has a lot of application for real world systems.
c)Dynamo is highly scalable systems following always on ,eventually consistent model.
d)Systems is optimized for 99.9th percentile not average.
Always writable do better customer experience
Cons :
a)ACID properties compromised.
b)There is no security in system and all the nodes assumed to be trusted.
c)In order to make vector clock scalable author truncates oldest value
resulting in reconciliation problems which author ignores saying its not
surfaced in production.
d)Dynamo considers heterogeneity as one of the design principles but in
the experimental part they have carried out their experiment on all
homogeneous machines.
No comments:
Post a Comment