Keyun Cheng

Reading Notes: TKDE’23 PartitionChain

Title: PartitionChain: A Scalable and Reliable Data Storage Strategy for Permissioned Blockchain

Conference (TKDE’23): Link

Journal (): Link

Summary

This work proposes PartitionChain to address BFT-Store’s problem, where the computation overhead and network transmission overhead are high for blockchain initialization and re-encoding. In addition, it enhances security, as BFT-Store does not implement the security check (e.g., measuring malicious behaviors).

Main Contributions

Details

For this paper, I merely focus on the storage overhead reduction

• Data partitioning: PartitionChain applies inner-block coding, where each block needs to be divided into pieces and perform encoding.
  • The benefit: for decoding, it only needs to retrieve the pieces for the block itself, rather than other blocks.
  • The drawback: decoding is required for ALL blocks.

• Protocol: also PBFT + RS code

• Encoding: similar to BFT-Store, but the coding is performed on a per-block basis; it also applies CLAS(Certificate Aggregate Signature) to enhance security for BFT.

• Decoding: it’s also similar to BFT-Store. It needs election of a leader that’s likely to be honest, and it will decode the block by retrieving valid pieces (from both honest and malicious nodes) of original blocks.

• Scaling: in my opinion, the scaling process for data re-encoding is the same as BFT-Store. The difference is the data verification process (in other words, PartitionChain verifies the data integrity of the original block before scaling).

• Benefits of PartitionChain over BFT-Store
  • Computation overhead reduction for decoding: as it limits the size of the chunk, and no cross-coding is performed, then the amount of data for decoding a block is definitely reduced.
  • The re-initialization overhead of PartitionChain is reduced from BFT-Store

Strength

• PartitionChain out-performs over BFT-Store mainly from the security aspect

• The re-initialization overhead of PartitionChain is reduced from BFT-Store, as some nodes doe snot need to re-encode all data

Weakness

• I think the evaluation is not sufficient, and the comparisions between PartitionChain and BFT-Store are not fair. Especially, the encoding/decoding throughput is on a per-block basis, that is to say, cross-coding must be of higher network transmission overhead inner-block coding. However, the drawback of inner-block coding, is that whenever a block is generated, coding needs to be performed immediately, which also incur significant computation/network overhead. Instead, cross-coding is performed on a per-batch basis, where blocks are grouped together and coded