347
© The Author(s) 2020
S. Scargall, Programming Persistent Memory, https://doi.org/10.1007/978-1-4842-4932-1_18
CHAPTER 18
Remote Persistent
Memory
This chapter provides an overview of how persistent memory – and the programming
concepts that were introduced in this book – can be used to access persistent memory
located in remote servers connected via a network. A combination of TCP/IP or RDMA
network hardware and software running on the servers containing persistent memory
provide direct remote access to persistent memory.
Having remote direct memory access via a high-performance network connection is
a critical use case for most cloud deployments of persistent memory. Typically, in highavailability or highly redundant use cases, data written locally to persistent memory is
not considered reliable until it has been replicated to two or more remote persistent
memory devices on separate remote servers. We describe this push model design later in
this chapter.
While it is certainly possible to use existing TCP/IP networking infrastructures to
remotely access the persistent memory, this chapter focuses on the use of remote direct
memory access (RDMA). Direct memory access (DMA) allows data movement on a
platform to be off-loaded to a hardware DMA engine that moves that data on behalf of
the CPU, freeing it to do other important tasks during the data move. RDMA applies the
same concept and enables data movement between remote servers to occur without the
CPU on either server having to be directly involved.
This chapter’s content and the PMDK librpmem remote persistent memory library
that is discussed assume the use of RDMA, but the concepts discussed here can apply to
other networking interconnects and protocols.
© The Author(s) 2020
S. Scargall, Programming Persistent Memory, https://doi.org/10.1007/978-1-4842-4932-1_18
CHAPTER 18
Remote Persistent
Memory
This chapter provides an overview of how persistent memory – and the programming
concepts that were introduced in this book – can be used to access persistent memory
located in remote servers connected via a network. A combination of TCP/IP or RDMA
network hardware and software running on the servers containing persistent memory
provide direct remote access to persistent memory.
Having remote direct memory access via a high-performance network connection is
a critical use case for most cloud deployments of persistent memory. Typically, in highavailability or highly redundant use cases, data written locally to persistent memory is
not considered reliable until it has been replicated to two or more remote persistent
memory devices on separate remote servers. We describe this push model design later in
this chapter.
While it is certainly possible to use existing TCP/IP networking infrastructures to
remotely access the persistent memory, this chapter focuses on the use of remote direct
memory access (RDMA). Direct memory access (DMA) allows data movement on a
platform to be off-loaded to a hardware DMA engine that moves that data on behalf of
the CPU, freeing it to do other important tasks during the data move. RDMA applies the
same concept and enables data movement between remote servers to occur without the
CPU on either server having to be directly involved.
This chapter’s content and the PMDK librpmem remote persistent memory library
that is discussed assume the use of RDMA, but the concepts discussed here can apply to
other networking interconnects and protocols.