Optimizing Data-intensive Systems in Disaggregated DataCenters with TELEPORT @Sigmod22

  1. The paper claimed that as brought in [2], MESI virtual cache is a better way of managing disaggregating memory. So they design an RDMA Daemon to do the MESI virtual cache things. For [2] is to leverage the DRAM TLB, that giving tag that can be VA or PA, so that better for communicating between cores. For this paper, they implemented a syscall called pushdown to emulate the memory sharing with different memory spaces with memory order assured, coherence protocol inside, and fast PA/VA translation between 2 memory spaces.
  2. The system is built upon LegoOS, which has no innovation if they implemented on Linux since there are so many daemon implementations out there.
  3. The daemon RPC messages are received and handled in a FIFO order, no transaction, no crash consistency, no atomicity.
  4. Concurrent page fault handling


  1. https://www.cis.upenn.edu/~sga001/papers/teleport-sigmod22.pdf
  2. https://research.cs.wisc.edu/multifacet/papers/isca13_direct_segment.pdf