Cloud4Home — Enhancing Data Services with @Home Clouds
Summary:
There is a trend of continued increasing capability of end devices, and also in the public cloud computing space. End devices like smart phones are releasing more often than the server systems which is used to drive the cloud services space. The downside on only pure @home capabilities is that it cannot utilize the large storage and computation capabilites of the public cloud. Similarly using only @datacenter capabilites might cause a lot of overhead when the end-devices are only accessing the internet, due to network latency. Thus combined use of @home and @datacenter is worth exploring. Cloud4Home project discussed in this paper exploits the benefits of both @home and @datacenter operations. It provides a dynamically varying sets of devices that cooperate to provide end users with seamless storage, access and data manipulation services.
Strengths of the Paper:
1) Caching of key-value entries onto intermediate hops on each request's path greatly improves the availability and reliability.
2) The project is guided by strong design principles : 1) Fungibility, 2) Augmentation, 3) Guided active management, 4) Automation and Independence. These principles are important because there may be downtimes in remote cloud or end-devices but the services should be up and running despite that. Also, resources on both the environments should be virtualized such that they can be used interchangibly.
3) System can easily solve the privacy and security concerns which few countries have for the data being collected from local users.
4) The paper has evaluated benefits of the system based on various services like home surveillance, media conversion etc. These experiments show VStore++ overheads for inter-node transfers and effect of increase in object size, Tradeoffs in data placement, the flexibility offered by VStore++ in combining object manipulation functions with storage.
5) Store operation has strong control over where data is stored. This provides better ability to meet privacy and performance demands.
Weaknesses of the Paper:
1) The outcome comparision of Cloud4Home with pure cloud and 'at the edge' is based on the quality in service provided, which may not be the only aspect the users look for. Scalability is also an important aspect and the system performs poorly with respect to it.
2) Global indexing of resource availability by VStore++ may not be a scalable solution.
3) There hasn't been any comparision with any exciting or related systems.
Discussion Points:
1) Is there a better approach for updating the metadata entries on intermediate hops?
2) Cost of providing richer access control policies for better privacy?
3) Is there a way to combine clusters of Cloud4Home system? Will it solve the scalability issue?
4) Is dedup and compression an option while transferring the blocks across Cloud4Home system? If yes, what are the benefits?
Summary:
There is a trend of continued increasing capability of end devices, and also in the public cloud computing space. End devices like smart phones are releasing more often than the server systems which is used to drive the cloud services space. The downside on only pure @home capabilities is that it cannot utilize the large storage and computation capabilites of the public cloud. Similarly using only @datacenter capabilites might cause a lot of overhead when the end-devices are only accessing the internet, due to network latency. Thus combined use of @home and @datacenter is worth exploring. Cloud4Home project discussed in this paper exploits the benefits of both @home and @datacenter operations. It provides a dynamically varying sets of devices that cooperate to provide end users with seamless storage, access and data manipulation services.
Strengths of the Paper:
1) Caching of key-value entries onto intermediate hops on each request's path greatly improves the availability and reliability.
2) The project is guided by strong design principles : 1) Fungibility, 2) Augmentation, 3) Guided active management, 4) Automation and Independence. These principles are important because there may be downtimes in remote cloud or end-devices but the services should be up and running despite that. Also, resources on both the environments should be virtualized such that they can be used interchangibly.
3) System can easily solve the privacy and security concerns which few countries have for the data being collected from local users.
4) The paper has evaluated benefits of the system based on various services like home surveillance, media conversion etc. These experiments show VStore++ overheads for inter-node transfers and effect of increase in object size, Tradeoffs in data placement, the flexibility offered by VStore++ in combining object manipulation functions with storage.
5) Store operation has strong control over where data is stored. This provides better ability to meet privacy and performance demands.
Weaknesses of the Paper:
1) The outcome comparision of Cloud4Home with pure cloud and 'at the edge' is based on the quality in service provided, which may not be the only aspect the users look for. Scalability is also an important aspect and the system performs poorly with respect to it.
2) Global indexing of resource availability by VStore++ may not be a scalable solution.
3) There hasn't been any comparision with any exciting or related systems.
Discussion Points:
1) Is there a better approach for updating the metadata entries on intermediate hops?
2) Cost of providing richer access control policies for better privacy?
3) Is there a way to combine clusters of Cloud4Home system? Will it solve the scalability issue?
4) Is dedup and compression an option while transferring the blocks across Cloud4Home system? If yes, what are the benefits?
Good blog. What about cloud4home leads you to be concerned about scalability? Your discussion points are fine but feel a little generic. For example, #4 could be asked of any system transmitting data OR is there a specific reason why you think dedup/compression might be needed and/or valuable given the context of this paper?
ReplyDeleteFor #4: Since VStore++ exposes object based file system handler and dedup is not that suitable if most of the objects stored are different. So, if somehow we can improve the benefits of object level dedup we can reduce a lot on network bandwidth usage. What I think is that edge devices have different characteristics and hence will have some variance while object creation, which will lead to different objects most of the time(I am not very sure about this, please correct me if I am wrong).
ReplyDeleteFor #2 and #3: We could have discussed what changes are required based on the implementation of the system discussed. These 2 points are kind of open ended but we can get some fresh/new thought process out of it.
ReplyDelete