The Fabric architecture and transaction pipeline should be very familiar to the readers of this book by now. It is a complex distributed system and its performance depends on many factors, ranging from the architecture of the application interacting with Fabric to consensus implementation, transaction size, block size, Fabric network size, as well as capability of the underlying hardware and physical network medium.

At the time of writing this book, performance measurements reveal that Fabric can yield a throughput of several thousand transactions per second (https://arxiv.org/abs/1801.10228.). The caveat our readers need to keep in mind is that these measurements were carried out using chaincodes that performed very simple operations, and using application and network configurations that may not represent a typical production blockchain network. Fabric performance is bound to the specific use case and the underlying hardware. For example, performance on IBM Z systems exceeds other platforms due to optimized Go compilers leveraging hardware acceleration capabilities such as for cryptographic algorithms and others. Good performance depends on the availability of sufficient resources and proper configuration; we will discuss configuration at length later in this section.