R&D Projects

[scott]

Above all I'd like to keep atomic focused on emerging trends, obviously we're focused on keeping our core products here in stable/enterprise development cycles but I don't want to lose sight of emerging technology.

To that end I want to point out some emerging projects that I feel we've under-researched here based on some of your input just to get the discussions going:

* nginx, dayo has been the champion of this project here for well over a year and in testing it showed a considerable promise. It might be able to replace apache entirely, it does or at the very least act as a significant HTTP accelerator. This has been implemented for plesk 9.5 environments in the atomic-accelerator package

* mysql 5.5, I think everyone including me has touched this. Its in atomic-testing now, the core questions here are what problems does it cause if we adopt it? Is this a major achievement, or just a version bump?

* Drizzle, highland bought this up first. Its a huge shift from the sql and even nosql worlds of database design. Is this an emerging trend? Or is it a flash in the pan? More importantly, what needs to happen for everyone to answer that question.

* mod_ruid2 & ITK/Apache 2.2.19.(Ikkk, antiochinteractive, and hostingguy) Everything else before this has been to some extent revolutionary... ITK and RUID2 are evolutionary changes to apache that could eliminate the need for solutions like fastcgi. We cover this in the plesk irc channel all the time (did you know we had an irc channel? ##plesk on irc.freenode.net) Is fastcgi good enough? Do we need to focus more on this project? If so what needs to happen to bring it up to enterprise level?

* atop, bruen mentioned this the other day, and I'd never heard of it until then (so kudos! This is why I wrote this post). It has a kernel plugin which is unique in all other performance measuring tools. This could be the next generation of tools like top.

Above all Atomic is about taking new cutting edge concepts and bringing them to everyone. Integration is part of the program, but more often than not we rely on you all to bring the new projects and prototypes to everyone's attention. What are we missing?

So if you have a new project, or implementation or just general improvement please share it here. Everything is on the table, from apache modules to kernel RPMS (3.0 anyone?). Above all I want to keep the process here transparent. Maybe we should keep enterprise packages in atomic and break R&D out into another channel... or conversely maybe atomic should be for the researchers and we create something for the enterprise class users so they can opt in.

[Highland]

The big change in MySQL 5.5 was multi-core support. So it's definitely not just a version bump. And Oracle just announced 5.6 so expect 5.1 to go EOL before the end of the year.

Drizzle may not kill MySQL single handedly but the number of MySQL forks is exploding (thanks, in no small part, to a broad dislike of Oracle and numerous former MySQL employees who jumped ship). In some places it's causing drama. I still like Drizzle's simplified approach to MySQL (only a core with everything else being a plugin) but it might not be for everyone. What is clear is that MySQL is splintering with lots of projects clamoring to replace it. There's Drizzle, SkySQL, Percona and MariaDB, just to name a few. I picked Drizzle out of that pack because they went back to the basics, instead of just trying to "fix what's wrong with MySQL" (which all MariaDB and Percona are).

I don't know that FCGI will go anywhere. The smaller footprint makes it ideal for smaller virtual servers, regardless of the ITK stuff.

When I attended a PHP conference in April, I was amazed at the adoption of nginx. Apache is clearly no longer a given in hosting. The other HTTP server gaining ground is lighthttpd.

[scott]

So on the subject of drizzle, do you know of any applications that can use it? I'd like to set it up here in the QA system to see what its capable of. Aside from looking at the way it gets built, I havent even downloaded it yet.

[Highland]

In theory, any application that runs on MySQL should run on Drizzle. I'd try a vanilla Wordpress install and see. I haven't had time to play with it but I was able to get their repo set up and get it installed.

[scott]

So you'd use the same mysql_connect() API to talk to it in a PHP example?

[breun]

Parallels already support MySQL 5.5 with Plesk 10. You can select it in the autoinstaller and it'll install it for you. I haven't tested it myself yet though.

[Highland]

So you'd use the same mysql_connect() API to talk to it in a PHP example?

AFAIK it should work (it's designed to work with the same mysql protocols or else nobody would even consider using). I don't know that it would work with mysqlnd, tho.
http://wiki.drizzle.org/MySQL_Differences
http://azhari.harahap.us/2011/04/connec ... hpmyadmin/

I hadn't looked but there is a guide by one of the Drizzle devs for installing Wordpress
http://www.linuxjedi.co.uk/?p=172

[RS_Jelle]

The big change in MySQL 5.5 was multi-core support. So it's definitely not just a version bump. And Oracle just announced 5.6 so expect 5.1 to go EOL before the end of the year.

It's more than that. They also replaced MyISAM with InnoDB as the default storage engine and InnoDB performance was improved. Better performance is always major to me

[scott]

Have you had a chance to use the atomic-testing build? If so how is it working out?

[hostingguy]

I hadn't looked but there is a guide by one of the Drizzle devs for installing Wordpress
http://www.linuxjedi.co.uk/?p=172

^^

An attempt has been made to create a plugin to use Drizzle, but unfortunately it has side-effects such as modifying your content if you happen to blog about anything related to MySQL or Drizzle.

So the first rule of drizzle is that you dont talk about drizzle?
IMO drizzle will never kill mysql - if anything does it would be oracle or mysql imploding on itself. Drizzle will probably never move mainstream so long as it takes a lot of effort for a sys admin to install and configure and make sure there are plugins for everything, etc. Unless its one of those things like mysql where you set it and virtually forget about it, its market adoption rate is virutally 0 - somelike akin to postgresql.


ITK is nice, as it would allow everyone to run php - but its still a ways off from major consumption. For us it seemed to break anything that wasnt PHP as DSO - FCGI didnt work, tomcat didnt work, asp, etc. The way to get arround that is to only turn it on for select sites - which in my mind makes it worthless as then you could just turn them to cgi or fcgi instead and have no problems.


I would like to see a more expanded integration of LVE and virtual containers into the kernel so that things like mod_hostinglimits and limiting virtual host consumption would become more mainstream. Its hard to imagine that shared web hosting has been arround this long with out a more mainstream method of controlling memory, cpu, file usage per virtual host. I am a little biased here though as we currently are not using yoru kernel so that we could move to a solution that did provide LVE support (which I must say has dramatically increased server performance, uptime and reliability during all times of the day) but that leaves us slighly exposed to some of the protections your kernel does offer.


On that same note I would like to see LVE expanded into vixie-cron so that the same protections can be put in place to cron jobs - whether its a customer job or the Plesk daily jobs so that a cron job cant take down your box as well.

[premierhosting]

Nginx as an apache replacement would be fantastic. If it could drop in on top of a Plesk install even better.

[mikeshinn]

Nginx as an apache replacement would be fantastic. If it could drop in on top of a Plesk install even better.

Keep in mind that there is no WAF module for nginx, so if you use nginx and you want to be able to block web attacks you will need to put a WAF in front of it.

[premierhosting]

https://www.atomicorp.com/forum/viewtop ... f=1&t=5018

[scott]

Oh on the LVE front, we're coordinating directly with Cloudlinux on the 2.6.32.x kernels. We've got the point on this, because Cloudlinux needs to use parts of openvz and we were already working on getting openvz into shape for newer kernels.

Maybe we should break this thread out into different themes (or forums even?). Id like to spend more time focusing on getting those leading edge projects up to something more testable. Hostingguy brought up exactly the kind of problem I want to focus the repo on solving, Getting pain-in-the-ass to integrate projects into an easily installable form.

[dayo]

Nice to see ART keeping up to speed on things.