Issue #126
After 5 years, I'm out of the serverless compute cult, Scale to Zero Problem, Networking and Distributed Systems, Pathways: Asynchronous Distributed Dataflow for ML
Simplicity does not precede complexity, but follows it.
— Alan Perlis
Note to readers - There will no issue next weekend.
Posts
After 5 years, I'm out of the serverless compute cult
In a cult, brainwashing is done so gradually, people have no idea it is going on. I feel like this has happened across the board with so many developers; many don’t even realize they are clouded. In my case, I took the serverless marketing and hype hook, line, and sinker for the first half of my serverless journey. After working with several companies small and large, I have been continually disappointed as our projects grew. - #BrentMitchell #Dev
The Black Swan Events in Distributed Systems
So, EBS servers use a buddy system where every block of data is stored on 2 different servers for reliability. When a server loses connection to its buddy, it assumes that its buddy has crashed and now it has the one and only copy of customer data. Now, this server quickly tries to find a new buddy in the hope of replicating customer data. This process is called re-mirroring.
In this scenario, since there was a big network outage, when the network outage was resolved, there was a big group of EBS servers, all looking for a buddy to mirror the customer data.
Maybe at this point you can already guess what happened next, as they called it a re-mirroring storm. -#AshutoshNarang #Medium
Cloud is expensive. One reason for this is that a typical application instance sits idly waiting for inbound requests. Yet even though it is idle, it is incurring cloud charges. What if there was a way to scale down your application when it was not being used? What if it could be scaled to the point where no instance were even running? This is the idea behind “scale-to-zero” designs. -#MattButcher #fermyon
Software Architecture and Design InfoQ Trends Report—April 2021
Dapr and the Open Application Model are two ways to make building distributed systems easier, and it will be interesting to watch how they are adopted in the future.
The pendulum seems to be coming to rest, after swinging to extremes between monoliths and microservices. As a result, architects are relying on well-established patterns and designs that focus on high cohesion and low coupling, regardless of the underlying technology.
In fully-remote work environments, architects are finding new ways to communicate with their teams, and finding replacements for the water cooler chats which were useful for gathering knowledge.
The next generation of GraphQL features, notably GraphQL Federation and GraphQL Microservices, are showing where to go next after companies have strong adoption of GraphQL.
Video
HC32-S6: Networking and Distributed Systems
Programmable Ethernet Switch - Anurag Agrawal and Changhoon Kim
Distributed Services Architecture - Francis Matus
The Fungible DPU™: A New Category of Microprocessor for the Data-Centric Era - Pradeep Sindhu
Fungible High-density Multi-tenant Bare-metal Cloud with Memory Expansion SoC and Power Management - Justin Song and Xiantao Zhang
Paper
Pathways: Asynchronous Distributed Dataflow for ML
PATHWAYS makes use of a novel asynchronous distributed dataflow design that lets the control plane execute in parallel despite dependencies in the data plane. This design, with careful engineering, allows PATHWAYS to adopt a single-controller model that makes it easier to express complex new parallelism patterns. We demonstrate that PATHWAYS can achieve performance parity (∼ 100% accelerator utilization) with state-of-the-art systems when running SPMD computations over 2048 TPUs, while also delivering throughput comparable to the SPMD case for Transformer models that are pipelined across 16 stages, or sharded across two islands of accelerators connected over a data center network.
#PaulBarham #AakankshaChowdhery #JeffDean #SanjayGhemawat #StevenHand #DanHurt #MichaelIsard #HyeontaekLim #RuomingPang #SudipRoy #BrennanSaeta #ParkerSchuh #RyanSepassi #LaurentElShafey #ChandramohanA.Thekkath #YonghuiWu