CODEWITHSUNDEEP
hyperledger-fabrichyperledger
tungnt185
tungnt185

Reputation: 51

Why Hyperledger Fabric’s performance is so bad in our deployment benchmark?

We use Hyperledger Fabric and Composer to build a system with a channel and 3 peers (8GB, 8CPU) but we faced 2 problems like below:

Performance is very low: 10 tps. Is it the limit of the composer or our fault in implementation?

Size is very large, it's more than 20 times compare to SQL Server. With 1000 records in SQL we've used 40MB but with Hyperledger system as above (3 peer) is 700MB (I think if it's linear the size of Hyperledger should be = 40*4 = 160MB). Is this size is normal or our fault? What's the best practices in size optimization?

Thanks in advances

Upvotes: 0

Views: 1138

Answers (2)

Glenn
Glenn

Reputation: 8032

I conducted some similar testing. I took an existing microservice and replaced the MySql, Redis, and Cassandra parts with API calls to a business network built with Hyperledger Composer. The storage capacity for the original microservice was 750 times more efficient than CouchDB under Fabric. You can read all about this experiment in this article on Evaluating Hyperledger Composer.

Upvotes: 1

yacovm
yacovm

Reputation: 5140

Performance is very low: 10 tps. Is it the limit of the composer or our fault in implementation?

I don't know about composer but it's definitely not the limit of Fabric. More data about your deployment would be helpful.

Size is very large, it's more than 20 times compare to SQL Server. With 1000 records in SQL we've used 40MB but with Hyperledger system as above (3 peer) is 700MB (I think if it's linear the size of Hyperledger should be = 40*4 = 160MB).

A transaction with a single endorsement weighs about 3K. 1000*3K is about 3MB... something here is off ;) How many blocks and transactions do you have?

Upvotes: 1

Related Questions