1. 8

  2. 6

    One nice feature of FIFO’s is that if you write less than PIPE_BUF bytes, the write is atomic. This is not necessarily so useful in the 1 producer multiple consumer case, but in multiple producers one consumer case it means the file lock can be removed alltogether. Unfortunately, PIPE_BUF is small on OS’s like FreeBSD (512 bytes, I think), but Linux is around 8kb, I think.

    The other thing to note is that the FIFO is not doing fan-out. Each consumer is destructive to the FIFO.

    All-in-all, I think FIFOs are an underused tool in a lot of software. The kernel is pretty efficient at shuffling data around, the cost is a a syscall to read/write to a FIFO, but it’s a nice way to connect components and maintain process boundaries. Unless someone is doing really high throughput/low latency work where disrupting the CPU cache is a serious problem, using FIFOs to connect components will probably have no negative effect on performance.

    1. 2

      Thank you for your comment. FIFOs can be a bit limited in terms of atomic writes, and in they are indeed not recommended for multiple consumers. I’ve extended the post to add more details about some of the limitations.

    2. 2

      @wsdookadr since you made a blog post, a few comments about how it works would be nice for the non bash/*nix trivial stuff. E.g. how it works in general, what is ‘200>$LOCKFILE’, what does flock do, etc.