Disqus - Latest Comments for mtrencseni

Re: Can Foundational Physics Be Saved?

Marton Trencseni — Sun, 16 Dec 2018 02:47:35 -0000

One thing I wish would happen: create a nice, asthetic, smooth way for a smart student to go from the two-slit experiment to QED/QCD [in 2-3 years], while having clarity what the input parameters are in the model (=needs to be measured), what is predicted by the model (=this comes out and is measurable), where an approximation is being made, what the current spacetime model is assumed to be, etc. Also, a mathematician should be able to look at this "journey" and approve that it's not too handwavy.

Given that there's "not much to do", it's weird that there isn't a 1% group somewhere out there who thinks this is worthwhile.

A good example (but probably too mathy) is the work of Tamas Matolcsi: `Spacetime without reference frames` and `Ordinary thermodynamics`.

https://www.amazon.com/s/re...

Re: Bytepawn - New Logitech M705 Marathon Mouse

Marton Trencseni — Thu, 07 Oct 2010 15:30:14 -0000

Buy the Anywhere, I bought it to replace the Marathon, and it's great:

http://bytepawn.com/2010/10...

Re: Bytepawn - Making Mysql safe for transactional usage

Marton Trencseni — Fri, 30 Jul 2010 01:30:35 -0000

You should reproduce the problem by forming a hypothesis what the hack was and taking the SQL queries your system performs and executing them on the console (several consoles) per you hypothesis, and see if you can perform the hack. Then you can decide whether you want to fix this is at the isolation levels (if that's even a fix) or change the application code. Best of luck!

Re: Scalien | Blog - Keyspace 1.6.2 now featuring ASCII Art

Marton Trencseni — Mon, 17 May 2010 17:08:21 -0000

Log rotation is currently not supported.

Re: Scalien | Blog - Keyspace 1.6.2 now featuring ASCII Art

Marton Trencseni — Mon, 10 May 2010 09:58:33 -0000

I added the Perl docs:

http://scalien.com/document...

Re: Scalien | Blog - Keyspace Roadmap

Marton Trencseni — Wed, 31 Mar 2010 12:46:55 -0000

+ maybe use Git

Re: Scalien | Blog - Keyspace Roadmap

Marton Trencseni — Wed, 31 Mar 2010 11:21:17 -0000

Actually, the Ruby client is done.

Re: Scalien | Blog - Keyspace 1.6, Python libs, new docs

Marton Trencseni — Wed, 24 Mar 2010 11:23:10 -0000

With SWIG we can port to a bunch of other languages as well.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Tue, 01 Dec 2009 15:14:45 -0000

One that uses your hard disk?

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Tue, 01 Dec 2009 15:03:44 -0000

Yes, Redis is fast as it's doing everything in-memory. I never said it was slow. I put it into that category because it's limited by system memory, hence you can't use economically to store lots of data, which tends to come up when "scalability" is an issue. (I don't want to pay the monthly fee for a server with 64GB of RAM.)

I may be wrong though, Redis is seeing lots of use, the author has hit a sweetspot.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 17:21:23 -0000

If you read the article you will notice that I'm not discounting anything. I simply point out that there are web developers who use "NoSQL" databases such as CouchDB to get rid of Mysql.

Also, I don't think the SQL guru portion of its user community is what makes CouchDB so successful. I believe it's the web developer community, who give it all the great publicity. There's nothing wrong with that.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 16:43:14 -0000

The article is not meant to be insulting. I wasn't making a general statement about all CouchDB users. I was pointing out that I believe CouchDB owes much of its success to its usefulness as a lightweight tool used by web programmers who weren't using the "advanced" (note the quotes..) features of SQL databases, and that this use-case is worth examining on its own. I'm not implying web programmers who are not SQL gurus are lesser people.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 16:34:32 -0000

There are no programming languages that magically scale at no cost. "There is no silver bullet."

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 16:30:40 -0000

You're right. I wrote Hadoop because HBase is under the Hadoop project. Perhaps I should have been more specific.

Actually, if you read the paper, Bigtable is a key-value store underneath, with an API that makes it appear to be a table; the columns of rows are serialized into (row:column) keys.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 14:43:49 -0000

But "overall" speed still matters. I am not trying to "diss" CouchDB in this article; CouchDB obviously has many more users than Keyspace because of its excellent features, and in this respect Keyspace has a lot to learn from CouchDB. Granted CouchDB had a 2 year headstart, but that's a compliment to Damien Katz, who had good insights early.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 14:17:35 -0000

If your data set or load is large enough that "scalability" comes up, you usually need "speed" to go along with it.

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 13:21:11 -0000

Thanks!

Re: Bytepawn - The Confused World of "NoSQL"

Marton Trencseni — Mon, 30 Nov 2009 12:46:42 -0000

I haven't tried MongoDB, but based on its docs it's more "for web designer" than for large-scale operations. I'm fairly knowledgable about replication, and I'm skeptical about their approach. Their sharding is in alpha, so it's too early to tell.

Re: Bytepawn - Scalable Web Architectures and Application State

Marton Trencseni — Mon, 27 Jul 2009 17:53:01 -0000

If you do an amazon search for "scalable" you'll get a bunch of results, but I can't tell you whether these books are any good..

Re: Bytepawn - Scalable Web Architectures and Application State

Marton Trencseni — Mon, 27 Jul 2009 12:28:22 -0000

What I wrote about here is an edge-case of web-development. I don't think there are many places where even web development is taught, mostly because it's a fairly simple area [in terms of what you can learn at a school] after a general CS curriculum. Also, these are practical issues which only come up if you work on real-life projects, whereas courses usually concentrate on theoretical subjects. These are issues which used to be talked about in trade journals, such as the C/C++ User's Journal; today trade journals have been replaced by blogs. So you should 1. go to school for a general CS curriculum 2. read good books 3. read bood blogs 4. read good research-level papers (see below).

If you're willing to dig a little bit deeper, learn about databases, operating systems and distributed systems in general, and you will get insights which will serve you well in all other areas. I have a 'Readings in Distributed Systems' section here on bytepawn at http://bytepawn.com/reading... but note that these research-level papers pretty much assume familiarity with database concepts.

Re: Bytepawn - Scalable Web Architectures and Application State

Marton Trencseni — Sat, 18 Jul 2009 15:52:03 -0000

We're working on a pure PHP driver (it's certainly prossible), we'll have an announcement when it's done!

Re: Scalien | Blog - Scaliens! Scaliens! Scaliens!

Marton Trencseni — Sat, 18 Jul 2009 08:43:16 -0000

Don't worry about Paxos patent issues...

Re: Bytepawn - Scalable Web Architectures and Application State

Marton Trencseni — Fri, 17 Jul 2009 05:05:18 -0000

I wrote this post based on some recent "consulting" work. Based on this, most shops still run a basic LAMP stack and put everything in Mysql. They've never heard of Memcached. Rails is not representative of the "web industry".

Re: Bytepawn - Scalable Web Architectures and Application State

Marton Trencseni — Fri, 17 Jul 2009 04:17:32 -0000

Application state is something that can be re-initialized from the database upon restart.

Think of this: suppose you have infinite memory and you store *everything* in memory. This is good as long as you don't have to restart your server / software, because then you'd loose everything, including user registrations and whatnot. So you obviously store some things in the database. Now suppose your restart your server: data you can restore from the database by some initialization script and then keep in memory until the next restart is, roughly speaking, application state.

This is of course a question of design decisions. Suppose you are running a web-based chat system. Things like user registrations should go into the database. Permanent chat rooms should go into the database. Information about which user is logged in, what chat room he's in, etc. this is application state: if you restart your server, they'll have to re-login and re-join the chatrooms. (Note that your code is probably written in a way where this happens anyways.) Actual chat messages should go in the database, because it's nice to have a persistent transcript, but they're not super-important, so you should save up a set of chat messages in memory and write to the database once every minute. If the server goes down, some messages may be lost, but it's no big deal, and will keep your server load down.

Re: Bytepawn - Thoughts on Yahoo's PNUTS distributed database

Marton Trencseni — Wed, 18 Feb 2009 15:35:05 -0000

I printed out the FATLease paper a while ago but I haven't read it yet. I'll post a reply once I have..