This week the team double down on the topics of in-memory computing and persistent memory. Chris and Martin talk to Charles Fan, CEO at MemVerge about Big Memory and using persistent memory technology (specifically Optane) to supplement system DRAM.
DRAM is expensive and as capacities scale linearly, the price of memory increases exponentially. Systems are limited by maximum addressable memory per socket. Persistent Memory in the form of Intel Optane provides the capability to massively increase the virtual memory footprint, using a combination of DRAM and Optane DIMMs. How is that memory managed – this is where MemVerge comes in.
MemVerge uses a feature called LD_PRELOAD to intercept memory management calls and direct requests to the MemVerge software. This means MemVerge can manage applications sitting across real DRAM and Optane. It also means the company can introduce neat tricks like memory snapshots and application rollbacks. This is a great conversation and serves as a great introduction in what we can expect of in-memory computing in the future. Find more on MemVerge at https://www.memverge.com/
Elapsed Time: 00:39:10
- 00:00:00 – Intros
- 00:02:15 – Real-time computing is driving in-memory computing
- 00:03:00 – DRAM is not scalable and expensive
- 00:05:30 – Applications can hit a cliff edge when in-memory capacity is exhausted
- 00:07:45 – Persistent Memory (like Optane) is byte-addressable
- 00:09:30 – Memory capacities scale linearly and costs increase exponentially
- 00:11:00 – How are systems going to balance DRAM and PMEM access in a single server?
- 00:13:00 – Application rewrites are not popular…
- 00:15:30 – How is PMEM identified by the O/S?
- 00:16:45 – MemVerge implements “software-defined memory” – LD_PRELOAD
- 00:19:30 – How is MemVerge technology implemented?
- 00:21:30 – What algorithms are used to manage DRAM and PMEM?
- 00:24:45 – Test figures show performance can be better than DRAM
- 00:28:00 – How do you use the persistence of PMEM effectively?
- 00:30:00 – MemVerge offers application checkpointing – memory snapshots
- 00:33:00 – Snapshots in memory and on disk are co-ordinated
- 00:36:45 – Persistent Memory will be a game-changer for in-memory computing
- 00:38:00 – Wrap Up
Related Podcasts & Blogs
- #169 – In-Memory Computing and Apache Ignite
- #36 – The Persistence of Memory with Rob Peglar
- #159 – Introduction to MRAM with Joe O’Hare from Everspin
- Persistent Memory in the Data Centre
- What are Storage Class Memory (SCM) and Persistent Memory (PM)?
Copyright (c) 2016-2020 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #747k.