a1exwang / bartender

A distributed NVRAM-based memory allocator. A related programming language "adonis" is at https://github.com/a1exwang/adonis-lang

bartender

A high performance distributed guaranteed failure-atomic NVRAM-based memory allocator.

What is "bartender"?

"bartender" is a distributed NVRAM-based memory allocator, with a traditional C-like memory management interface(malloc/free) and can be deployed on a cluster of host machines connected by InfiniBand or Ethernet to provide a transparent memory allocation API for the clients.

Features

• NVRAM-based
  • Guaranteed failure-atomic
  • Failure recovery
  • Transaction
• Distributed clusters
  • Transparent
• Concurrency
  • Cache locality
  • Contention
• Auto defragmentation

Challenges

• For NVRAM
  1. Memory leak on NVRAM.
  2. Data consistency and its overhead
• For distributed memory allocators
  1. Consistency
  2. Latency and throughput
  3. Locality
  4. Fragmentation
• For concurrent systems
  1. Consistency
  2. Contention
  3. Memory blowup

Requirements

For development

• Hardware
  • PC
• Software
  • VirtualBox
  • rustc

For production

TODO

Setup

Single node

• This repository
• rust
• gcc

Multiple nodes on single machine

TODO

Usage

#include <bartender.h>

struct btd_config btd_config;
btd_config.single_node = 1;

struct btd_cluster *p_cluster = btd_cluster_create(&btd_config);
const char *s = "123";
char *p_data = btd_malloc(p_cluster, 100);
btd_memcpy(p_data, s, strlen(s));
btd_free(p_data);

Benchmark

• nvm_malloc benchmark https://github.com/IMCG/nvm-malloc/tree/master/benchmark

Design

TODO

TODOs

1. Malaku

   • Features
     • NVRAM-based memory allocator
     • Support for more familiar C-like malloc/free interface
     • Decrease latency by lazily persisting non-core metadata
     • Post-failure GC for data consistency

2. Hoard

   • Hoard GitHub
   • Address following issues
     • Contention
     • False Sharing
     • Memory Blowup

3. Fast Cluster Communication

   • Accelio
   • or write a custom IB/RDMA-based high speed RPC/data transfer library

4. Distributed Memory Management

• e.g. maybe CEPH, ZFS or GFS for consistency, performance, fragmentation

1. Benchmark tools or methods

• nvm_malloc benchmarks

1. IB, NVRAM Emulator

• Use RAM disk + mmap to emulate NVRAM
• Use local RAM to emulate remote memory or Ethernet-based RDMA?

1. GitBooks

• Try GitBooks for our documentation

1. Easy LaTeX editor or transformer

• maybe pandoc + vim?

1. New programming language with NVM variable support?

References

• [1] Bhandari K, Chakrabarti D R, Boehm H. Makalu: fast recoverable allocation of non-volatile memory[J]. Acm Sigplan Notices, 2016, 51(10):677-694.
• [2] Mellanox, http://www.mellanox.com/related-docs/prod_software/RDMA_Aware_Programming_user_manual.pdf
• [3] Berger E D, Mckinley K S, Blumofe R D, et al. Hoard: a scalable memory allocator for multithreaded applications[C]// International Conference on Architectural Support for Programming Languages and Operating Systems. ACM, 2000:117-128.
• [4] Volos H, Tack A J, Swift M M. Mnemosyne:lightweight persistent memory[J]. Acm Sigarch Computer Architecture News, 2011, 47(4):91-104.
• [5] Coburn J, Caulfield A M, Akel A, et al. NV-Heaps: making persistent objects fast and safe with next-generation, non-volatile memories[C]// Sixteenth International Conference on Architectural Support for Programming Languages and Operating Systems. ACM, 2011:105-118.
• [6] Christopher Mitchell, Yifeng Geng, Jinyang Li. Using one-sided RDMA reads to build a fast, CPU-efficient key-value store[C]// Usenix Conference on Technical Conference. USENIX Association, 2013:103-114.
• [7] Kalia A, Kaminsky M, Andersen D G. Using RDMA efficiently for key-value services[C]// ACM Conference on SIGCOMM. ACM, 2014:295-306.
• [8] Megalloc: Fast Distributed Memory Allocator for NVM-Based Cluster
• [9] BDWGC, A garbage collector used in Makalu, https://github.com/ivmai/bdwgc.git
• [10] pmem pmem, Intel's NVM Programming library
• [11] pmem valgrind Enhanced Valgrind for Persistent Memory
• [12] NVL-C A extended C for NVM