[Links of the Day] 02/01/2020 : Video and Slides for Networking @scale and NeurIPS 2019 + AWS batch job monitoring

• NeurIPS 2019 : All slides and video from the biggest AI / ML conference this year.
• Networking @scale 2019 : All video from Facebook Networking at scale conference
• batchiepatchie : A really cool project that allows you to monitor, gather metrics and display useful information about your AWS batch jobs.

[Links of the Day] 31/12/2019 : C++ STL iterators for GO, Hotchips 2019 and Rust implementation of Graph based Nearest Neighbour search

• iter : Go implementation of C++ STL iterators and algorithms. I really need to look into C++ again. Been a really long time since I touched it.
• HotChips 2019 : All videos from talks at this year HotChips conference ( 31st of its kind!)
• Granne : Graph-based approximate nearest neighbour search in Rust, I really like Annoy from Spotify but this Rust implementation seems attractive enough that I will give it a spin. 

[Links of the Day] 28/11/2019 : Metrics Stream Processing Framework, SOPS'19, Distributed Transaction K/V

• Mantis : netflix metric stream processing framework. This seems to be a really powerful platform for realtime monitoring and especially contextual alerting and alerting on logs. What's impressive is the throughput it can churn through events...  [github]
• SOPS'19 : SOPS papers are in free open access, a lot of papers on security and formal verification of distributed this year. [Murat review Day0, Day1, Day2]
• Tikv : Distributed transactional key-value database.  [Github]

[Links of the Day] 15/10/2019 : k8s python operator, Strangeloop notes, Machine learning UI framework

• Kopf : Kubernetes Operator Pythonic Framework— is a framework and a library to make Kubernetes operators development easier, just in few lines of Python code.
• Strange loop notes : some quick and insightful notes on this year strange loop.
• StreamLit : app framework specifically for Machine Learning and Data Science teams. It helps you build quick UI around your ML models and get the result out quick to the end user. [website]

[Links of the Day] 21/02/2019: Database Internals, Fosdem 2019 Videos, Cloud Programming Simplified

• Database Internals ; excellent series delving into the internal mechanism and algorithm of modern and not so modern database systems.
• Fosdem 2019 Videos: Fosdem 2019 conference video start to filter through on the interweb
• Cloud Programming Simplified: A Berkeley View on Serverless Computing paper which gives a quick history of cloud computing, including an accounting of the predictions of the 2009 Berkeley View of Cloud Computing paper, explains the motivation for serverless computing, describes applications that stretch the current limits of serverless, and then lists obstacles and research opportunities required for serverless computing to fulfil its full potential.

[Links of the Day] 21/05/2018 : Automation and Make, FoundationDB, Usenix NSDI18

• Automation and Make : this is a really good description of best practice for Makefile and automation. 
• FoundationDB : Apple open source it's distributed DB system, another contender enters the fray. With Spanner on google cloud, CockroachDB and now FoundationDB. The Highly resilient distributed transactional system start to reach widespread usage.  [Github]
• Usenix NSDI 2018 Notes: a very good overview of NSDI conference, and naturally the morning paper is currently doing a more in-depth analysis of the main papers. [day 2&3]

[Links of the Day] 07/05/2018 : Cryptocurrency Consensus Algorithms , Fast18 conference, Google 2 real world project translation

• A Hitchhiker’s Guide to Consensus Algorithms: this post provides a quick and easy way to understand the classification of the various cryptocurrency consensus models. It's a gentle introduction to the concept of proof of work vs proof of stakes vs proof of authority vs ... Well, you got it many many more algorithm.
• Notes from FAST18 : a very good overview of the Storage conference. What is becoming obvious over the years is that a lot of the actual practical implementation of novel distributed storage solution is directly pushed into Ceph. Ceph is poised to become the defacto default private storage solution even if it has a long way to go in term of manageability and automation. I think it stems from the preconception that a lot of operations need a storage admin person. But the projects like Helm are helping it get there.
• xg2xg : a practical translation table of internal google tech and similar technology available to those that do not work in the chocolate factory. It is a very good list of production-ready project that can be leveraged in many devops (and non-devops) environment.

[Links of the Day] 11/01/2018 : Two machine learning conference NIPS 2017 & Robot Learning CoRL 2017 , CS Paper ML detector can still be fooled too easily

• Nips : This conference is considered one of the biggest events in ML\DNN Research community. Here are two sets of notes from the conference by ‎Olga Liakhovich and by David Abel. These are two fairly long article but worth a read. Looks like fairness and bias is one of the big topics of the moment. Also, I like how ML is compared to alchemy. The current approach is extremely fragile, tailor-made and not fully understood. Too often machine learning tools are considered black box where you shove in data at one end and get a result on the other. 
• Conference on Robot Learning (CoRL) : robot and machine learning are converging at an aggressive pace. It is rather impressive how all these different aspects of computer science are clicking together and with each small improvement in each domain lead to an overall jump in robotic capability. 
• Adversarial Examples that Fool Detectors : last but not least, common machine learning classifiers are still way too fragile and can be easily fooled. With the boom in use of ML technique everywhere. This can become really quickly a problem in the near future. 

[Links of the Day] 02/11/2017 : Probabilistic programming Library, Better than Word2Vec , Serverless conf

• Edward : Turing complete deep probabilistic programming software library. Probabilistic computing is going to be at the forefront of the next big AI improvement wave. [software]
• Better than Word2Vec : first word2vec is great and there is already so many pre-built libraries out there that it should be your number 1 go to approach. Then if you want to develop custom word embedding library, as the blog post explains, SVD might be a better approach. 
• ServerlessConf NYC 2017 : a good summary of New York serverless conference  2017. 

[Links of the Day] 24/10/2017 : W3C Payment request API, Xen Unicore, Kafka Summit

• Payment Request API : Interesting to see that now all browser implement the w3c payment api. One step closer to more fluid banking service consumption. 
• Xen Unicore : Proposal to build a core set of functionality and library allowing users to build unikernels targeted at specific applications without requiring the time-consuming, expert work that building such a unikernel requires today.
• Kafka Summit : Videos & Slides from the San Francisco 2017 are now available.

[Links of the Day] 19/10/2017 : Distroless minimal container images, Strangeloop and CppCon 2017

• Distroless : a toolkit for creating container images which contain only your application and its runtime dependencies. No package shells, managers, any other programs. This is really awesome as you always end up with a lot of clutter in your layer. too often you see apt-get update ; apt-get upgrade in your Dockerfile. Or you could simply move to Golang and enjoy a From scratch environment! [talk] [video]
• Strangeloop: a very good review of the excellent strange loop conference. Every time I watch some of these talks, I tell myself I should get into functional programming .. Then life ( and three kids under 6 ) takes over. [slides] [videos]
• CppCon : another dev conference, this time C++. Some interesting talks, especially the includeOS one. Which provides a C++ microkernel functionality by simply including a single header #include ! A lot of in-depth and technical talks, give it a look if you are a C++ dev. [slides] [videos]

[Links of the Day] 05/09/2017 : Patent Surviving Alice, Hot Cloud 17 conference papers

• 7 Post-Alice Patent Cases That Survived 101 Rejections : Alice US supreme court decision started a slaughter in the US patent office regarding IT related patent: 8400 applications dropped and 60k+ rejected. While courts invalidated the vast majority of patent litigation. However, it seems that there is a way to survive the onslaught, and it's quite simple. You just need your patent to satisfy the following criteria: novelty, enablement, non-obvious, and last but not least useful. it seems like a no brainer, but it seems that the USPTO allowed itself to be flooded by sub par applications that gamed the system. Not to mention that the agency financially gained from such practice to some extent also. 
• Hot cloud 17 : hot cloud conference just finished, here is a selection of interesting paper
• JavaScript for extending low-latency in-memory key-value stores : Adrian Colyer takes a look at in memory javascript engine using RamCloud. RamCloud project is entering the use case phase of the research project,  eyeing the comercialisation. Sadly most solution put forward are extremely niche. The risk is for great people to be stuck in a zombie startup if they try to run with it. Taking separately, the tech that came out of the RamCloud project is amazing. However, the solution as a whole doesn't really have a great killer app or any potential beyond some niche market.  [paper]
• Towards Index-based Global Trading in Cloud Spot Markets :  the authors propose to use an index based prediction model rather than per spot instance in order to obtain greater reliability at lower cost.
• DAL: A Locality-Optimizing Distributed Shared Memory System :  Different take on the whole in memory K/V system, the authors aggressively move the data to the computation rather than offering remote access. This allows great data reuse. We used something similar in hecatonchire. However, there is a certain risk when you have a high level of churn or serial data access and local caching of data generate a high level of eviction, effectively doubling the bandwidth usage. 
• Leader or Majority: Why have one when you can have both? : raft is a great consensus protocol ( and easier to understand). However, the over reliance on the leader is the main bottleneck for scalability of operations. The authors ( from cockroachdb ) propose a quorum based read operations that allow alleviating the load on the leader while still retaining strong consistency. This allows them to improve write by 4x write perf and increase throughput by 60%. Which is quite impressive. 
• DCCast: Efficient Point to Multipoint Transfers Across Datacenters : the authors proposed an efficient multipoint data transfer protocol allowing greater efficiency and bandwidth usage.

[Links of the Day] 06/07/2017 : venture capital investment framework, Aftershock in complex systems, ISC workshop

• Picking Winners : the authors propose a framework for venture capital investment. However, I feel that there is a major flaw in the approach as it is designed to correlate previous success with future investment success. This assumes a high degree of repeatability with clear identifiable elements. The model can suffer from the purple cow effect where higher growth opportunities are located in underserved segments. As a result, quite often by following the model, you might suffer from diminishing returns.
• Aftershocks in a complex system : the authors look at the behaviour of a complex system ( currency market) after a catastrophic event. They discovered that often such system follows a similar pattern to the one exist in earthquake's aftershock. In extenso, after an initial catastrophic event, most systems suffer a series of gradually diminishing aftershock spaced in time. 
• ISC Workshop 2017 : all about container and optimisation for high-performance workloads.

[Links of the Day] 08/03/2016 : Intel blockchain, Fast17 conference and papers, AWS cloud formation devops tool

After a small hiatus, here is the return of the links of the day.

• Sawtooth Lake: Intel distributed ledger system. It uses an interesting security mechanism to deliver secure consensus. Sadly it relies on Intel proprietary hardware encryption modules to deliver this feature.
• Fast17: File and Storage technology Usenix conference happened last month. There were a couple of interesting papers but one picked my interest: Redundancy Does Not Imply Fault Tolerance:Analysis of Distributed Storage Reactions toSingle Errors and Corruptions. The authors look at single file system fault impact on Redis, ZooKeeper, Cassandra, Kafka, RethinkDB, MongoDB, LogCabin, and CockroachDB. Turns out most systems are not able to handle these type of faults very well. It seems that a single node persistency layer error can have an adversarial ripple effect as distributed system seems to have put way to much trust in the reliability of this layer. Sadly they lack tools for recovering from errors or corruption emerging from file systems.
• Stacker : remind 101 tools for creating and updating AWS formation stacks. Looks like an interesting alternative to terraform. 

[Links of the Day] 09/01/2017 : Incident Response process, Plain english Legal guide to start a startup, 33C3 videos

• Incident Response : pager duty incident response documentation. This is a very thorough and well documented process for handling incident, before during and after their occurrence. Probably not one size fit all by can be easily adapted to an company needs.
• Plain English legal guide on how to start a business : nice guide of the legal aspect of how to start a startup and the various options and pitfall associated with it. A little bit too US centric for my taste but still has some good insights. 
• 33C3 :Chaos Computer Congress videos are now available. Chris Hager gave a great overview of the different talks. As always a lot of diversity and challenging presentations.

[Links of the day] 10/12/2016 : Full stack fest 16, Product Dev handbook and Stealing Machine Learning Models

• Full Stack Fest 2016 : Playlist of the video of Full Stack fest 2016 conference
• Product Aikido : handbook for an organisation’s Product Development Group.
• Stealing Machine Learning Models : the authors of the papers propose a technique that analyse response of system using Machine learning via their API in order to extract the model used. And as a result allow the attacker to determine the best response for manipulating the system.

Notes on SNIA Storage Developer Conference 2016

This year SNIA Storage Developer Conference, chosen bits : 

• MarFS : scalable near-POSIX file system using object storage. What is really impressive is that MarFS is part of a 5 tiers storage system of the trinity project. Yes FIVE tiers, RAM -> BurstBuffer -> Lustre -> MarFS-> Tape. MarFs seats above Tape for long term archival and aim at providing storage persistence that span year(s) of usage. In comparison Lustre just above aim at keeping the data for weeks only. What bother me is the logic behind this approach as most Supercomputer system have a 5-6 year lifespan. This implies that the project usage will span multiple generation of systems. [Github]
• Hyperconverged Cache : It seems that Intel start to realize what we discovered years ago in the Hecatonchire project. Once you start to have near Ram performance, dis-aggregating and pooling your ressource becomes the natural next step for efficiency. And this is what they aim to achieve with a distributed storage cache system that would aggregate their 3dxpoint system across a cluster in order to deliver fast and coherent cache layer.  However without RDMA this approach seems a little bit pointless. The only things that seems to save them is that the cloud storage backend ( Ceph ) has a big enough latency gap they can exploit. 
• Erasure Code : Very good overview of modern erasure code and their trade-offs. As always no code are equal but not all use case are the same. 

Persistent Memory :  As storage shift away from HDD to Pmem , the number of talk around persistent memory exploded this year. The main focus seems to shift from pure NVM consumption to remote access model. 

• NVMe over fabric : two talk on the recent progress of NVMe over fabric. Nothing really new there, just that it seems that it will be the standard in remote storage access in the near future. [Mellanox] [Linux NVMf]
• RDMA :  It seems that Intel and other are aiming for direct Persistent memory access using RDMA, bypassing the NVMe stack. The idea is to eliminate the latency from the NVMe stack. However this require some change in the RDMA stack in order to guarantee persistence of data. 
• IOPMEM  : interesting work where the author propose to bypass CPU interaction between PCIe devices. Basically enabling DMA between NVM and other devices. It then allows RDMA NIC to directly talk to the NVM device on the same PCI switch. However it doesn't really explain what persistence guarantee are associated with the different operations.
• RDMA verbs extension : basically Mellanox propose to add a RDMA flush verbs that would mimic the CPU flush command . This operation would guarantee consistency and persistence of remote data. 
• PMoF : address the really difficult aspect of guaranteeing persistence and consistency of accessing persistent memory over fabric. Basically this talk describe all the nitty gritty detail to avoid losing/corrupting data during access over over fabric. This is what the RDMA flush verb will need to address but for the moment require a lot of manual operation. 

Last but not least we can see reference here and there to 3dXpoint from Intel however it seems that the company tuned down its marketing machine. Probably fearing some backlash because of the  continuous claw-back on claimed performance front. 

[Links of the day] 05/10/2016 : SQL Scan using NVMe to GPU using P2P DMA , Strange Loop 2016, Secure Time Service

• SSD-to-GPU Direct DMA : Interesting work where the author use p2p DMS to load data from NVMe to GPU. This bypass the RAM altogether. The objective is to accelerate PostgreSQL scan operation. This is really neat, but I am not sure that the SQL DB are the best choice of use case. I would have thought that columnar or K/V system would held better speedup potential because of the way the data is organised and processed. 
• Strange Loop : all the video of strange loop 2016 conference. Way too many good talk to mention them all. Just check it out. 
• RoughTime : secure time synchronization project. Everybody use time, but nobody mention clock attack. This project aim at alleviating this potential threat. So you know all your data is accurately fresh. 

[Links of the day] 23/09/2016 : Intel's 3dxpoint vanishing performance, VLDB16, Core to Core HW queue engine

• 3dxpoint performance evaporate : seems that Intel is heavily scaling back its xpoint NVM performance claim. From 1000x to 10x ( still good but a far cry from what was promised). It seems that Intel had to push the technology early in order to counter a potential acquisition of its partner, Micron, by a competitor. Announcing the technology surely propped the share price making an acquisition difficult. 
• VLDB : very large databases 2016 proceedings are out. Sadly its one big zip file and didn't have time to go through it.
• CAF : the authors propose a hardware core to core communication offloading engine. Providing an efficient queuing mechanism for transferring data between cores. I am not sure 100% of the value but the concept is interesting, let see if it catch on and if it can plays well in heterogeneous environment of today's datacenter. As core to core is slowly replaced with cored to GPU or core to FPGA or core to NVM.

[Links of the day] 21/09/2016 : NAS 16 , cloud optical interconnect, Netmiko

• NAS : 2016 networking, architecture and storage conference, selected papers: 
• CircularCache : storage wide cache system using virtual queue mechanism for load balancing usage and performance accross the cluster.
• Active Burst-Buffer : when storage is not fast enough you start to move your processing int he buffers to save time
• Emerging Optical interconnect Technology for the Cloud : Finisar presentation , the largest fiber optic transceivers provider in the world, on the trend in cloud interconnect technology. Like HPC and other system its all about power / bits / seconds. Power decrease while bandwidth needs to increase. What is interesting is the impact of the topology used on the fabric requirement ( HPC vs hyperscale datacenter). What is impressive is that aggregate bandwidth doubles every 3 years but that cost per Gbps is lower for higher channels counts at the same point in time.
• Netmiko : paramiko wrapper simplifying SSH connections to network device