A message-passing, thread-migrating operating system for a non-cache-coherent many-core architecture

The difference between emerging many-core architectures and their multi-core predecessors goes beyond just the number of cores incorporated on a chip. Current technologies for maintaining cache coherency are not scalable beyond a few dozen cores, and a lack of coherency presents a new paradigm for software developers to work with. While shared memory multithreading has been a viable and popular programming technique for multi-cores, the distributed nature of many-cores is more amenable to a model of share-nothing, message-passing threads. This model places different demands on a many-core operating system, and this thesis aims to understand and accommodate those demands.;We introduce Xipx, a port of the lightweight Embedded Xinu operating system to the many-core Intel Single-chip Cloud Computer (SCC). The SCC is a 48-core x86 architecture that lacks cache coherency. It features a fast mesh network-on-chip (NoC) and on-die "message passing buffers" to facilitate message-passing communications between cores. Running as a separate instance per core, Xipx takes advantage of this hardware in its implementation of a message-passing device. The device multiplexes the message passing hardware, thereby allowing multiple concurrent threads to share the hardware without interfering with each other. Xipx also features a limited framework for transparent thread migration. This achievement required fundamental modifications to the kernel, including incorporation of a new type of thread. Additionally, a minimalistic framework for bare-metal development on the SCC has been produced as a pragmatic offshoot of the work on Xipx.;This thesis discusses the design and implementation of the many-core extensions described above. While Xipx serves as a foundation for continued research on many-core operating systems, test results show good performance from both message passing and thread migration suggesting that, as it stands, Xipx is an effective platform for exploration of many-core development at the application level as well.