Mariano Guerra's Log (Publicaciones sobre PapersOfTheWeek)

Papers (and other things) of the LargeSpanOfTime II

Mariano Guerra — Thu, 31 Aug 2017 13:15:48 GMT

OK, the title is getting fuzzier and fuzzier, but I decided to condense some things I've been reading here.

Papers:

Bringing the Web up to Speed with WebAssembly:

I like compilers, and their implementations, so I've been following WebAssembly, this is a good place to look at.

A blog post by google made the rounds lately with people saying that google was saying that they beat the CAP Theorem, so I went to the source. The conclusion is interesting:

Spanner reasonably claims to be an “effectively CA” system despite operating over a wide area, as it is
always consistent and achieves greater than 5 9s availability. As with Chubby, this combination is possible
in practice if you control the whole network, which is rare over the wide area. Even then, it requires
significant redundancy of network paths, architectural planning to manage correlated failures, and very
careful operations, especially for upgrades. Even then outages will occur, in which case Spanner chooses
consistency over availability.
Spanner uses two-phase commit to achieve serializability, but it uses TrueTime for external consistency,
consistent reads without locking, and consistent snapshots.

Bitcoin: A Peer-to-Peer Electronic Cash System:

Again, many people ranting and raving about bitcoin, blockchain and cryptocurrencies, what's better than go to the source, really readable paper.

CAP Twelve Years Later: How the “Rules” Have Changed:

I have a deja vu that I already read this paper, but just to be sure I read it again, interesting summary of the concepts and how they evolved over time.

LSM-trie: An LSM-tree-based Ultra-Large Key-Value Store for Small Data:

I wanted to read the LSM-tree paper and it seems I didn't look what I was clicking so instead I ended up reading the LSM-trie paper, which is really interesting and has an overview of the LSM-tree one, now I have to go and read that one too.

A prettier printer Philip Wadler:

In a previous post I mentioned that I read "The Design of a Pretty-printing Library" and I was expecting something else, well, this paper is a something else that I liked more.

Metaobject protocols: Why we want them and what else they can do:

Being an aspiring Smug Lisp Weenie I had to read this one, it's a nice paper and puts a name on some "patterns" that I've observed but couldn't describe clearly.

The Cube Data Model: A Conceptual Model and Algebra for On-Line Analytical Processing in Data Warehouses:

I've been thinking lately about the relation between Pivot Tables, Data Cubes and the things mentioned in the paper A Layered Grammar of Graphics so I started reading more about Data Cubes, I skimmed a couple papers that I forgot to register somewhere but this one was one I actually registered.

End-to-End Arguments in System Design:

Someone somewhere mentioned this paper so I went to look, it's a really good one, like the Metaobject protocol paper and other's I've read, this one is like a condensation of years of knowledge and experiences that are really interesting to read.

Books:

Object-Oriented Programming in the Beta Programming Language:

Interesting book about a really interesting (and different) object oriented programming language by the creators of Simula (aka the creators of object orientation), it explains an abstraction called "patterns" in which all other abstractions are expressed.

Project Oberon The Design of an Operating System and Compiler:

Another interesting book by Niklaus Wirth, creator of between others, Pascal, Modula and Oberon describing how to basically create computing from scratch.

I will note that I skimmed over the dense specification parts of those books since I wasn't trying to implement nor use them.

Reading:

Unconventional Programming Paradigms
Your wish is My Command Giving Users the Power to Instruct their Software
The AWK Programming Language
Mindstorms: Children, Computers And Powerful Ideas: All About Logo, How It Was Invented And How It Works by Seymour Papert

Papers this looong week: 11 (count books as papers because why not)

Papers so far: 54

Papers in queue: don't know

Papers of the LargeSpanOfTime I

Mariano Guerra — Tue, 15 Nov 2016 20:47:29 GMT

Welp, some day the experiment had to end, I stopped reading 5 papers a week because some books arrived and I read those instead and also because I was busy at work.

But that doesn't mean I didn't read papers at all, so here's a list of the ones I did read.

Cuneiform: A Functional Language for Large Scale Scientific Data Analysis

Seems useful in practice, was expecting something else from the title.

The Stratosphere platform for big data analytics

I remember reading a paper from what later became Apache Flink that I liked a lot, I was looking for that one and I found this one instead (stratosphere became flink), it was an interesting overview, would like to know how much of that is still in flink.

Orleans: Distributed Virtual Actors for Programmability and Scalability

Really good paper, I like how it's written and the idea and implementation.

HyParView: a membership protocol for reliable gossip-based broadcast

Epidemic Broadcast Trees

This too reviewed together because they are like bread and butter, I love both of them, highly recommended.

Large-Scale Peer-to-Peer Autonomic Monitoring

I won't lie to you, I don't remember much about this one, but given the authors it must be good :)

Stream Processing with a Spreadsheet

Object Spreadsheets: A New Computational Model for End-User Development of Data-Centric Web Applications

I was looking for ideas and inspiration when I read these two, I liked both, Object Spreadsheets being the most interesting aproach.

A Layered Grammar of Graphics

Great paper, on my top list, maybe because I love the topic :)

Virtual Time and Global States of Distributed Systems

A must read if interested in vector clocks, the non math parts are good, I don't enjoy reading theormes a lot (not their fault).

Papers this looong week: 10

Papers so far: 43

Papers in queue: don't want to count anymore

Papers of the Week VII

Mariano Guerra — Tue, 14 Jun 2016 09:24:56 GMT

Because nothing lasts forever and after a week half traveling and a busy one I managed to read 4 papers this week.

The first one was interesting but comes from an area I will describe as "let's bend relational databases to fit Event Stream Processing", which is not bad per se but has things like joins and being able to remember past events that make its scalability (at least in terms of memory) quite hard, also it never discuses distribution, which is ok for the field but not what I'm looking for.

Cayuga: A General Purpose Event Monitoring System

The interesting part about this one is the part where it introduces Visibly Pushdown Languages something that looks really interesting but I couldn't find an introduction for mere mortals, the descriptions are really dense an mathematical, which is ok but hard to learn for outsiders like me.

Another interesting point is the fact that it uses the XML Schema to optimize the generated VPA (Visibly Pushdown Automata) and that the implementation not only applies to XML but to any nested semistructured data.

High-Performance Complex Event Processing over XML Streams

The review of the next one will seem conflicting with my previous reviews, but this one had too much enfasis on the low level implementation details, not novel things and optimizations, just a lot of details, like the guys found the implementation really cool and wanted to share it with the world. Not a bad thing per se, but in this batch I was looking for abstractions, optimizations and distribution characteristics of stream processing, better if focused on distributed systems, and this one talked mainly about the DSL they build that compiles to C. It also sorts the streams, does multiple passes over the data, does lookahead in the stream and does a kind of "micro batches" which isn't what I was looking for.

Hancock: A Language for Extracting Signatures from Data Streams

The last one, I found the approach interesting, they seemed to try to push the purity of the approach (everything is a regular expression) which may have end up with a nice model (a thing I like) but by reading the code it doesn't seem to be really clear, at least for a OO/functional background, and I think less for non programmers. Maybe the syntax doesn't help and some other syntax would make things clearer, I don't know.

Other than that the approach is interesting and it made me think on some ways to define a stream processing language using mainly pattern matching.

EventScript: An Event-Processing Language Based on Regular Expressions with Actions

Papers this week: 4

Papers so far: 33

Papers in queue: 76

Papers of the Week VI

Mariano Guerra — Wed, 01 Jun 2016 08:45:44 GMT

Better late than never (even when I read all the papers last week) here is the sixth installment of Papers of the Week.

Starting next week I will try to write the reviews after I read the papers and not almost one week after when my memories are fuzzy :)

The fist one describes an implementation of out of order processing using punctuation, interesting in that it "applies" the concept of punctuation to building a streaming system and analyzes the result.

Out-of-Order Processing: A New Architecture for High- Performance Stream Systems

This one describes an implementation of a storage engine using LSM Trees and a compression technique.

Rose: Compressed, log-structured replication

You can read an overview of the next paper and find the link to it at acolyer's paper of the day: Holistic Configuration Management at Facebook, I copy the first paragraph here:

This paper gives a comprehensive description of the use cases, design,
implementation, and usage statistics of a suite of tools that manage
Facebook’s conﬁguration end-to-end, including the frontend products,
backend systems, and mobile apps.

It's a good overview of tools and techniques used to scale and standardize configuration management and how to avoid problems introduced by sloppy configuration management.

Holistic Configuration Management at Facebook

The next one is my favorite of the week, it defines a baseline by implementing solutions from other papers that introduce some parallelization strategy by implementing them in a simple single threaded way and benchmarking it against other solutions, then defined a "metric" that describes how many cores are required to match the single thread implementation, as many sites would tell you "the result will amaze you".

Scalability! But at what COST?

The last one for this week surprisingly brought me to the CRDT/Lasp/@cmeik land, when the title didn't seemed to imply that, the crazy fact is that I saw a talk about this paper at RICON 2015 and I didn't remembered the title :)

Some parts where hard for me since it's the first paper I read about CRDTs so I don't have the vocabulary and basic theory in place but it made me think on some interesting applications on the IoT and monitoring spaces.

Declarative, Sliding Window Aggregations for Computations at the Edge

Papers this week: 5

Papers so far: 29

Papers in queue: 82

Papers of the Week V

Mariano Guerra — Mon, 23 May 2016 21:00:07 GMT

I'm already late so let's go:

The first one, Discretized Streams, is the one I liked the most, it's about the the theory behind what became Spark Streaming, really interesting.

Discretized Streams: Fault-Tolerant Streaming Computation at Scale

The second one is interesting in its introduction of punctuation which it explains really well.

Exploiting Punctuation Semantics in Continuous Data Streams

Didn't liked this one too much, doesn't mean it's bad, just that sometimes the title gives me an idea on what it's about and when it's not I loose interest

Consistent Streaming Through Time: A Vision for Event Stream Processing

This one was interesting in its description of different types of windows and the definition of windows sematincs, I get the feeling that if I read it back in some not too distant future I will get a lot more out of it.

Semantics and Evaluation Techniques for Window Aggregates in Data Streams

More statistical than I tought it would be, but still I learned some things about random sampling. I guess it's one of those papers that are great if you are looking for a solution and this paper tells you what to implement.

Weighted Random Sampling over Data Streams

Papers this week: 5

Papers so far: 24

Papers in queue: 85 (I cleaned some duplicates and similar papers)

Papers of the Week IV

Mariano Guerra — Mon, 16 May 2016 14:18:48 GMT

Better late than never, and proving that I can count to 4, here we go with the 4th straight paper reading week.

MillWheel: Fault-Tolerant Stream Processing at Internet Scale

I liked the MillWheel paper, Google and Microsoft write really nice papers from the ones I've read.

Dryad: Distributed Data-Parallel Programs from Sequential Building Blocks

Didn't liked the dryad paper, was expecting something else.

PacificA: Replication in Log-Based Distributed Storage Systems

The PacificA paper is my favorite of the week.

Naiad: A Timely Dataflow System

About Naiad, I liked the idea about tracking distributed progress.

The Design of a Pretty-printing Library

I had big expectations for this paper, but it was too haskellish for my taste, I was expecting something else.

Papers this week: 5

Papers so far: 19

Papers in queue: 91

Papers of the Week III

Mariano Guerra — Tue, 10 May 2016 07:17:58 GMT

No, I didn't gave up, last week was short because of a holyday and a "bridge day" so I was riding my bike through the black forest, but I still read the papers I set up to read by cramming all of them in 3 days :)

First, a classic in distributed systems :)

Brewer’s Conjecture and the Feasibility of Consistent, Available, Partition-Tolerant Web Services

Papers of the Week II

Mariano Guerra — Sat, 30 Apr 2016 11:35:19 GMT

In my continuous attempt to see how far I can count in roman numerals here is the second week, still going, still 5 papers.

The one I liked the most was "The Dataflow Model..." mainly because it fired some ideas related to a problem I'm trying to solve.

The others were also good except "Reimplementing the Cedar File System Using Logging and Group Commit", mainly because it wasn't what I was expecting it would be.

Read this week: 5

Total read: 10

In the Queue: 61

Papers of The Week I

Mariano Guerra — Sun, 24 Apr 2016 20:40:47 GMT

This is an attempt to treat what I would call acolyer's syndrome which is the gilt felt by people that would like to read papers as often as Adrian Colyer but never do.

So I will blog the ones I read here to try to follow Jerry Seinfeld's Productivity Secret

I will blog weekly because I won't read a paper a day, but I will try to read around 4 or 5 papers a week if they are around 12~15 pages, if they are longer I will read less.

The initial topics are stream processing systems and distributed systems, I will follow the references that I find interesting to inform future papers.

I will also read papers that I find interesting as I go.

ok, without further ado, here are the ones I read this week.

Related to stream processing:

Clasics I wanted to read:

In queue: 33