Presentation: Rethinking Applications for the NVM Era
What You’ll Learn
- Approaches to handling persistent objects in DRAM
- How to handle failure in memory as if you were using a file system.
Abstract
Storage looks and feels like a block device hidden behind layers of abstractions - system calls, pagecaches, block device drivers and things we don't want to think about. This might change in the near future with the introduction of CPU accessible, byte addressable, persistent memory. How would your logging service, database or filesystem look in this world? Actually, would you need any of those or could a persistent C++ STL container be your storage service of choice?
What is your motivation for this talk?
So the focus of this talk is about how do you re-architect your software to take advantage of the advances of software today. Many developers treats the serialization of objects as sort of a separate process something that hangs off the side where the object is out of flat file or disk. But now if your object lives in persistent memory how do you manage that? How do you update that in a consistent way so it's you know it's just constant even through power failures and so on. So most of the focus is on talking about how we write our software in the future when our DRAM is persistent.
Who should come to your talk?
Somebody who writes C++ could but might be curious about what non-volatile memory might mean for them. So for instance maybe somebody who is developing a boost library and wants to use non-volatile memory and also wants to know what NVM is doing under the hood.
What can people come take away from this talk?
So I'm going to sort of start by talking about general principles. C had sort of an object layout in memory and if the the memory was persistent, how would you manipulate that in a failure atomically? What does it means if I'm inserting a node of a red/black tree and now pull power in between, how do I ensure I haven’t destroyed the black tree. So this has been tackled in the file system and database world. But I'm talking about a general principle that we're tackling this with non-volatile memory and I'm sort of going to move on to saying in a practical sense if you’re writing code today, what are the libraries available to let you do this in a more high level maybe having to take up all those details yourself.
What keeps you up at night?
Well as I look a lot of the hardware software interface so a lot of the tech issues I deal with is really coming from two angles. One is because Moore's Law is slowing down so CPUs basically you can not take the same code and speed up by a factor of two every 18 months. At the same time software developers cannot expect code to be sped up until they start taking advantage of new features in the CPU; because that's the only way you can get things to go faster. So really a lot of the stuff that keeps me awake at night is bridging this.
Similar Talks
License Compliance for Your Container Supply Chain
Open Source Engineer @VMware
Nisha Kumar
Observability in the SSC: Seeing Into Your Build System
Engineer @honeycombio
Ben Hartshorne
Evolution of Edge @Netflix
Engineering Leader @Netflix
Vasily Vlasov
Mistakes and Discoveries While Cultivating Ownership
Engineering Manager @Netflix in Cloud Infrastructure
Aaron Blohowiak
Optimizing Yourself: Neurodiversity in Tech
Consultant @Microsoft
Elizabeth Schneider
Monitoring and Tracing @Netflix Streaming Data Infrastructure
Architect & Engineer in Real Time Data Infrastructure Team @Netflix
Allen Wang
Future of Data Engineering
Distinguished Engineer @WePay
Chris Riccomini
Coding without Complexity
CEO/Cofounder @darklang
Ellen Chisa
Holistic EdTech & Diversity
Holistic Tech Coach @unlockacademy