RubyConf 2007 - Day 2 - "The Ruby VM Smackdown Begins"

As the sun rose over the hills of North Carolina, the quiet betrayed a seething fire blazing in the hearts of three development teams. "Three teams enter, one team leaves!" Yes, I am talking about the main event, people, the one, the only, Ruby VM Smackdown! The room was jam packed, the anticipation was high.

As good fortune would have it, I was sitting right next to the head of the team who was my personal favorite to win the day. Yes, I am referring to Evan Phoenix, leader of Team Rubinius along with co-conspirator Ryan Davis. You should have seen the look at their faces when the JRuby team showed the Rubinius RSpec test suite as part of their demo! But I am getting ahead of myself...

I like the underdog in a fight, when the underdog is tough, and under-estimated by their opponents. On one side, the resources of Microsoft, on the other, the full power of Sun. And then, THE LADS. If you don't want to see that one go down, you have no passion for the cutting edge of innovation. And if you do not know about the history of innovation in the software industry, you might miscalculate the odds. Gentlemen, to your corners... I want a clean fight!

MC Chad Fowler got up, and gave a brief intro to the show. "We are about to see what Matz called 'wonderous' last night," he said. And it was true.

First up, was John Lam of Microsoft with IronRuby armed with "State of IronRuby".

Next, Charles Nutter and Thomas Enebo of Sun with JRuby using "Ruby for the JVM" style.

Last, the challenger, Evan Phoenix with Rubinius wielding "Rubinius 1.0".

Wow. Bread and circuses, indeed!

I will be posting individually about each of the Ruby VM contestants, and will update the above links when I am done...

After all that excitement, the crowd was starved. We shuffled out to eat our lunches, and converse on all we had seen. From everyone I spoke with, the mood in the crowd was firmly in the Rubinius corner, with a close second to JRuby, with third place probably going to YARV. Hardly anyone I spoke with at RubyConf had any interest in using IronRuby at all!

That is to be expected from a conference like RubyConf, well known for its strong ABM (Anything But Microsoft) attitude, which is pretty unfair. That said, I think the uptake in the .NET community will be the measure of the success of IronRuby, not whether they can get the typical RubyConf attendee to switch back to Windows.

In the afternoon, I got to see some great sessions, from:

"Refactotum: Ruby" - Stuart Halloway
"Maximizing Productivity" - Eric Hodel
"Mac OS X Loves Ruby" - Laurent Sansonetti

I will be posting individually for each session when I get a chance, and will update the links here.

Once the afternoon's sessions had wrapped up, a large group of us overran a local BBQ joint named Mac's, with tasty BBQ, a great beer selection, and a bunch of Harleys in the parking lot. Despite our somewhat atypical appearance (nerd chic) for NC, we were welcomed warmly by the locals. There must've been 30+ people from RubyConf there, it was great!

We then returned at the perfect moment to hear Matz's keynote speech. And it was good. Really good. Matz is charming, and quite funny. His wit was so much better able to emerge in a forum where he was simply presenting, rather than the somewhat less successful "town hall" format that had been attempted the evening before. I have a separate posting with his remarks which will be up later:

RubyConf 2007 - Keynote by Matz

Once the keynote had ended, I just had to go up and shake the man's hand. I know it may be corny, but I really did suddenly appreciate what he really does, which is to inspire the community. People keep looking to Matz for all the answers, but the answers are up to us, my friends. No superheros, remember?

Afterwards, I want to the legendary RejectConf, held right next door. Don't ask me how I wedged my way in, but I did it. To everyone who's toes I stepped on: sorry! There was a fun series of lightning talk (5-10 min) presentations of pretty much brand-new stuff. The beauty of RejectConf is its appeal to the ADD'd-out techno-junkie: every few minutes something else. I will be posting a rundown of RejectConf when I get a chance here:

RubyConf 2007 - RejectConf

RejectConf was really a great highlight of RubyConf. Once they kicked us out, the entire crowd immediately migrated directly next door to the hotel bar where we drank beers and chatted loudly about techno-babble to our great enjoyment. I staggered up to my room, to rest up and prepare myself for Day 3.

Weighing In On The Upcoming Ruby VM Smackdown

Earlier today, I read with great interest the potentially inflammatory post where Ola Bini proclaims the Ruby VM wars already over, and basically divides the Ruby VM empire into two halves, with the center controlled by Rome, err, I mean Rubinius primitives.

I had already come to this conclusion a while ago. Not because I am smarter than anyone, FAR from it. Probably because I am more superficial, and easily impressed by meaningless characteristics, like how cool is the name Rubinius?

Seriously, back to something more erudite, like Ola's post. Notably missing from this future of Ruby utopia is any IronRuby/Ruby.NET vision. Is this a simple oversight, due to the pro-Sun anti-MS bias that any Java programmer has to have, or more likely knowing the source, is it because of the sadly glacial progress of Lam and his new friends, as compared to both Rubinius and especially JRuby?

Man, it must SUCK to work at Microsoft sometimes!

RubyConf 2007 => Ruby VM Smackdown

I just registered for RubyConf 2007, which will be held Nov. 2-4 in Charlotte, NC. If you want a more intimate, more hard-core Rubyist experience, than what RailsConf is turning into, RubyConf looks to be it.

There not being that many tickets available, you better register right now if you want to go. I registered without even looking at the agenda first... just filled out the registration form, and hoped it's was not too late.

Once I has nervously typed my credit card info, and gotten my email confirmation, I took a few breaths. and got around to checking out the schedule for the conference I had just committed to attending. I was not disappointed.

Lo and behold, the ultimate battle of the Ruby VM's will be taking place on Saturday, Nov. 3. Take a peek at what's in store:

9:00A - John Lam - State of IronRuby
10:00A - Charlie Nutter - JRuby: Ruby for the JVM
11:00A - Evan Phoenix - Rubinius 1.0

Wow! If you care at all about the future of Ruby as a language, not just as a vehicle for fevered dreams of Ruby on Rails-based world-domination, you can't help but get excited. The three ascendants to the throne of dynamic language glory, all duking it out in a back-to-back 3-hour frenzy. Yum!

Improving the language performance of Ruby is the big thing needed to push out all those other wanna-be next big languages into the dustbin of interesting languages that no one uses. I think Erlang is really neat, don't get me wrong, but one look at that syntax, and sugar is not what comes to mind!

Ruby/Microsoft?

Martin Fowler just posted on his bliki a little piece entitled RubyMicrosoft. As I read it, I felt a strange chill run down my spine. Perhaps a few words of explanation are needed...

At RailsConf recently I stood at the Thoughtworks booth for a moment chatting up Fowler, and embarrassing myself by asking for a photo with him (he doesn't pose for photos with people). Luckily, I was rescued by the arrival of Oli Bini, and an ensuing short discussion about JRuby. I expressed my admiration about how they had been "cleaning Microsoft's clock" as far as their much faster progress with a runtime for the Ruby language then Microsoft's efforts with the DLR.

Much to my surprise, Martin was actually paying attention to what we were saying (I mean to what I was saying, after all he knows and speaks with Ola), and chimed in about how Thoughtworks was working with Microsoft on language compatibility for IronRuby. I expressed a poorly articulated skepticism at the prospects.

Lo and behold, the RubyMicrosoft blog posting sums up rather well the half-formed mumblings I was trying to stammer out. Well, this is why he is a keynote speaker, and I am a dead programmer. However, impressed as I am at his intellect, I am not sure I share his optimism about Microsoft.

I certainly don't know if I am an "alpha geek" or anything like that. I'd rather be a jazz programmer than a rock star. But given my long history working with Microsoft technology, and then the seemingly contradictory fact that I was at both RailsConfs, I must really identify with the movement. Chalk it up as just one more developer who Microsoft is losing to the rebellion.

RailsConf 2007 - Day 2

After the incredible energy of RailsConf Day 1, Day 2 seemed to slow down the pace a little bit. That was good for me, as I could spend some time catching up with friends both old and new, and not just attending sessions like an information junky. First would come the morning keynote speeches from Cyndi Mitchell of ThoughtWorks, and headliner Tim Bray of Sun Microsystems, with a short interlude from Charles Nutter.

Cyndi Mitchell - Thoughtworks

Ms. Mitchell is UK Managing Director of ThoughtWorks. As such, she is a polished corporate representative, which is exactly the kind of emissary that Rails needs in big organizations. She had a few interesting points:

* 40% of new work at ThoughtWorks is coming from Ruby on Rails in the enterprise
* We need to reclaim the word "enterprise" back from meaning "bloated, inefficient, and expensive" to its original meaning of "entrepreneurial"
* The "real" enterprises are using Ruby on Rails to get stuff to market quickly
* ThoughtWorks is offering a complete Rails production stack called RubyWorks, and is will be providing paid 24/7 support for Rails.

Tim Bray - Sun Microsystems
After Cyndi, the irrepressible Tim Bray took the stage. Despite not really being that much into Java, I have been following Tim's blog for quite some time, as he is a smart thinker with plenty to say to the industry at large.

Tim thinks that the Ruby language is as important as Rails (I agree!!). With JRuby, Sun wants to sell lots of hardware to large enterprises that will be choosing Sun Hardware and Solaris as the preferred platform for their Rails applications. Sun is also doing lots of work on NetBeans to do better Rails development.

Tim then briefly brought up Charles Nutter to give him some Rails street cred...er, I mean to explain why Ruby developers should care about JRuby.

Charles made two main points:
* A whole different way to do Ruby dev...using tried and true Java tools that already exist
* Move into enterprise where Java is accepted but Rails is still not

Then back to Tim:
It is not CIO's who lead industry...it is people like us.

How to make money on free products
* first you get adoption
* then you get deployments
* then you can monetize at point of value - no large business will adopt any tech that is not supported

Big companies will either get a handle on web 2.0 or get disintermediated

He assumes Rails will succeed beyond our wildest dreams.

Java, php, and .NET will never go away

The network is the computer...and its a heterogeneous network. And JRuby is how we get everything to talk to each other.

Java really consist of three main parts: language/JVM/Java APIs.
JRuby lets Rails programmers access the latter two.

How do we get everything to talk to each other? REST

Etags are a way to solve the REST problem with different versions of the same data.

ATOM - Everything should have a publish button: cameras, phones, browsers, etc

Is REST all there is? SOAP... WS-* 36 specs, ~1000 pages of documentation.

WSIT - interop with MS WCF

Hot developer issues
*static vs dynamic languages
*maintenance
*tooling
*integration
*concurrency
*time to market
*scaling
*load balancing
*observability
*file IO
*DBMS
*Shared nothing architectures
*CPU
*concurrency
*debugging
*unit test
*functional programming?

Which matter the most?
*Conventional wisdom for each language
- Java: performance, tooling
- php: web scaling, ease of use
- Rails - time to market, maintainability
*Tim says: The most important are Time to market, and maintainability

Extra Action Marching Band
After the speeches, I went to a session I didn't find that interesting at all, and so in the interest of not offending anyone else too badly, I shall let it remain anonymous...you know who you were. But near the end, a massive blast of sound came from the entrance to the conference center, and then proceeded past the sessions rooms toward the main hall.

It was the Extra Action Marching Band, an amazing avant-garde performance troupe. If San Francisco (which is where they are from) had a love-child with New Orleans, it would be something like this band.

Despite an almost-incident with the conference center security staff, they were able to finish their set. The music and performance were quite incredible, showing a high level of musicality as well as serious creativity. The bullhorn with a delay pedal attached was really amazing sounding...I have to make one like that for my harmonica.

After this unique musical interlude, it was back to the conference sessions...thamks to Tim Bray for the photograph.


Xen And The Art of Deployment
Ezra Zygmuntowicz of Engine Yard, Merb, and BackgrounDRb fame gave my favorite presentation of the day. The session was packed, with both information and people. If you wanted information about scalability, at least from the perspective of the stack, then this was the high point of the entire conference. Here are copious notes of what he said:

Serving up Rails right now is all about Mongrel

Mongral uses a Ragel state machine for validation

Why is mongrel better:
*HTTP is well known protocol
*Mongrel easy to setup and use
*Transparent wire protocol

Thread safety
*Rails not thread safe
*1 request at a time per mongrel

New dog, old tricks
*Wide # of options to load balance front end
*pen, pound, HA-proxy
*Lightspeed can serve static files
*Apache 2.2/mod_proxy_balancer can do same

Each current solution has problems

Nginx
*serious performance
*small resource footprint
*no leaks under heavy load
*killer rewrite and proxy mods
*author is cool and responsive
*Nginx + Mongrel
*This is THE stack
*apache for mod_dav_svn
*flexible config allows both serving static and dynamic requests
*fast!!!
*Nginx cannot do upload progress due to buffering
*no connection rate limiting for proxy modules yet

The future for nginx
*mod_rewite going away
*replace with http_script_module
*embed NekoVM directly into nginx

Perfect simple stack
*Linux
*Nginx
*Mongrel (mongrel_cluster)
*Monit

Swiftiply - the cool new solution

Evented Mongrel
*hot patch
*remove ruby treeads
*replace with eventmachine event loop
*mongrel single threaded
*big IO speed increase
*stand up to higher load

How is single threaded faster?
*Ruby green threads are slow
*mutex locks are expensive
*one process can only do so much IO
*event driven means running in a tight loop and firing callbacks
*without context switching between threads, single process has less overhead

Mongrel vs evented mongrel - performance stats
*1 user - 1K more req/sec
*100 users - 758 req/sec vs. 3700+

Swiftiply proxy
*event drive proxy, small memory footprint
*faster than ha-proxy

How is swiftiply different
*In a normal proxy, you need to config each back end server in advance before starting proxy
*swiftiply is a client/server architecture, where new back end servers register with proxy
*keeps persistent connection between proxy and back end servers
*auto-configuration of proxy

This means you can start/stop as many mongrels as you want, and they get configured into proxy, opening the door to auto-scaling

The Zen of Xen
*monolithic linux VS modular lnux
*old model: one box running all services
*new: specific virtual machines setup for specific purpose

We all want to modularize application why not modularize servers

What about more than 1 box?

Old school - more boxes
*mysql box
*another box for other services

New school
*add another xen box
*pick resource intensive parts, and more to another VM
*each VM only runs one thing, but well
*target performance problems
*scale better

Advanced clustering
*boot off dom0 off of USB drives
*SAN for all Xen domU
*red hat clustering suite for fencing and cluster options
*GFS for 100% posix file sys
*hardware load balancer, or dedicated box running Ultra Money or LVS

Create a fabric of compute and storage
*If a node fails, just swap in another
*Move hot VM's to less loaded machines
*Deploy app on 1 node, then bounce mongrels with GFS
*Fragment and page cache across nodes
*Scale from 1 or 2 VMs to as many as traffic requires and then back down again as traffic subsides

RAM RAM RAM
*most rails apps are RAM bound before they become CPU bound
*Avg mongrel serves 70-120 mMP PER mongrel
*RMagick is a typical RAM hog
*:include on active record also very bad performance
*95% of rails apps will leak memory at point or another

Use Rails plugin leakhouse to find leaks

Rails eats DB resources for breakfast

Most people's production apps hav no database indexes! Learn where and when to apply indexes.

ActiveRecord insulates developers so much to the point of massive inefficiencies.

You will need to denormalize data and use custom SQL statements to get performance

Other Tips
*do not use file system sessions
*script/runner is inefficient. do not load rails into bgprocesses.
*Use raw MySQL and Ruby without Rails if you can
*DO NOT use script/runner to pocess incoming email. Run deamon in loop and poll svr with net/pop3 or net/imap

Rails is great for 80/20 rule, but you are on your own for last 20%

Learn how to write custom mongrel handlers

When is optimization NOT premature?

Ruby is plenty fast, its Rails that is on slow side

Cache, cache, cache espcially static HTML files

Do not store full ActiveRecord object in session, just store hash to recreate it

Parting thoughts
*do not take anyone's word as gospel
*test and benchmark for yourself
*trust but verify

The Rest Of The Day
After Ezra'a session he was swarmed like a rock star. I know, cause I was standing nearby having a really nice chat with Zed Shaw the creator of Mongrel and Evan Phoenix, the creator of the Rubini.us Ruby runtime I was mentioning yesterday. You go, Ezra!

Later on, I got to hang out at the Engine Yard booth with Ezra for a few minutes, where Mike Brown and some other brilliant Kiwi whose name I didn't catch were chatting him up. I learned even more about hard-core clustering from listening to those guys go off...all I can say is I'm glad I have hosting at Engine Yard!

I wanted to go to Dr. Nic's RejectConf, but so many people told me they planned on going that I figured I wouldn't get in. I couldn't take the prospect of being rejected twice at the same conference, so I made other plans which turned out to be really fun.

After dinner with the same group of RailsConf friends from last year, with more delicious Portland beer, and excellent conversation, I headed to the BoF sessions, to hear about Amazon's EC2 service. I really wanted to learn more hard-core details about it, but so many people were just hearing about it for the first time, that the session was dominated by introductory questions. Oh, well!

I was about ready to turn in. Lucky for me, Rails Machine was having their party at my hotel. Bradley and company are really smart, cool people, and also offer a fantastic service, so I was glad to hang out with them for a few minutes. The most fun was jamming on harmonica with Joey deVilla the AccordianGuy. After hearing him last year, I had brought a couple of my harps with me just in case, and we finally had a chance to bust out a couple of songs together. That was really fun! Thanks, Joey. If anyone has any photos or video, please post it somewhere...if not, well, you had to be there.

The Ruby.NET Interview From Down Under

An interview with Dr. Wayne Kelly has been put up on Channel 9, Microsoft's video interview site. Dr. Kelly is head of the Ruby.NET compiler project at the Queensland University of Technology in Australia. As always, I am eager to see progress on the .NET side of the Ruby language wars. I drank my coffee, and put on my phones for the 15 minute interview.

After watching the video, I must say I am a little underwhelmed. Not to disparage the fine work of the good doctor on this worthy project, far from it. But I was not impressed by his comments about his lack of familiarity with the Ruby language itself. "If I was going to do any real Ruby programming, I would have to have the pickaxe book at my side to help me with the syntax, " says Dr. Kelly midway thru the interview.

Compare that to a zealous Ruby language fan like Ola Bini with the JRuby team. And I say this as a long time detractor of Java! How DARE they flaunt their superiority.

There were a couple of interesting points, one of which was the trouble with implementing the Ruby.NET compiler by directly matching Ruby language methods to CLR methods. The collection of CLR methods for an assembly being set at compile time, vs. Ruby class and instance methods which are a very fluid and wonderful thing and can be created at any time.

It was also curious that he said they are not using any of the .NET techniques being pioneered by the IronPython team, like use of dynamic delegates and Lightweight Code Generation. I was really expecting that they would be. At least I would be, if I were them.

Dr. Kelly was apologetic over his team's missing their self-imposed monthly delivery of updated code. He was so understated, that by the end of the interview I got the feeling that "under promise and over deliver" wasn't just cliche, but it was this guy's personal mantra. I hope I got the correct impression.

Let's see some bits, boys! Keep those updates coming...the race isn't over yet.

JRuby Has Won The Rails Race vs. .NET

So now it is official. The latest build from the JRuby team is now running Ruby on Rails in the JRuby runtime. Great work from that team! The Sunsters up top must be gloating right now.

And Microsoft is still MIA. Despite some hopeful signs of life, the silence is deafening from Ray Ozzie and the .NET camp (is that a band?). MS had better have some damn strong announcement at MIX vis-à-vis Ruby.NET or some top secret John Lam project if they want to stay in the game.

In the meantime, I still think that YARV will be the runtime for me, based on sheer performance alone. Not to mention that is now part of the Ruby SVN tree.

Even eWeek Is Falling For Ruby

Even that stanch old "enterprisey" publication eWeek has been giving some love to Ruby and Ruby on Rails. Today's most recent article is entitled "Ruby, Ruby, When Will You Be Mine?". Was this supposed to run on Valentine's Day? I'm definitely feeling the love in the title.

Anyhow, the author is all over the JRuby project. Messrs. Nutter et. al. have been building up the energy at the same time as the code base. Microsoft is notably receiving less coverage, no doubt due to providing no visibility as to what they are up to. I almost hate to give Sun props over Microsoft, after spending so many years arguing the opposite, but Sun is allowing the JRuby team to keep up the enthusiasm in the community. MS is giving the appearance of stagnation, even if they are just as active with development of Ruby.NET/RubyCLR/etc. For example, the Ruby.NET team is saying "This is a preliminary beta release. It is knowingly incomplete and contains many bugs. We therefore ask that you do not submit bug reports at this stage." Talk about the Wow...not. Meanwhile the JRuby team is involving the community in the entire dev process just like a real open source project does.

Based on the initial performance numbers from Antonio Cangiano, I'm probably not really looking at either one of those options for a production application any time soon. Can you say YARV?