Handling Flash Crowds From Your Garage

slashdotmsiriv writes "This paper from Microsoft Research describes the issues and tradeoffs a typical garage innovator encounters when building low-cost, scalable Internet services. The paper is a more formal analysis of the problems encountered and solutions employed a few months back when Animoto, with its new Facebook app, had to scale by a factor of 10 in 3 days. In addition, the article offers an overview of the current state of utility computing (S3, EC2, etc.) and of the most common strategies for building scalable Internet services."

Not Flash Mobs?

Here I was picturing a bunch of people showing up in your garage for seemingly no reason. Still interesting to see how they handled the massive increase!

Re:

And here I was thinking that it was about handling a crowd flashing for you in front of your garage...

Re:

I thought it was something to with actionscript.

Re:Not Flash Mobs?

Question: So how DO you handle flash crowds?

MS Answer: We don't. We'll force them to install Silverlight.

Where are the pix?

Since I can't see any pix in the area near the bottom called "Figure: DNS servers fail over very quickly when an upstream server fails" - does that mean that the flash crowd called "SLASHDOT" has taken down this part of the article called "Handling Flash Crowds..." ?

I mean sheesh! They even mention slashdot!

Re:Not Flash Mobs?

And here I was thinking that it was about handling a crowd flashing for you in front of your garage...

You must be non-American, because in Soviet America you flash the crowd.

Misleading pretense

"Our innovator may get only one shot at widespread publicity. If and when that happens, tens of thousands of people will visit her site. But a flash crowd is notoriously fickle; "

The "researchers" offer a strange view of how the market works. If the idea is good then surely the site will enjoy numerous opportunities for growth and referral every time a happy user recommends it to a friend. A good, innovative idea will not be sunk by one underprovisioned flash crowd.

Re:

I think you overestimate the attention span of the type of people who compulsively install Facebook apps.

Re:Misleading pretense

The first crowd is a different class of user from the general public. It's a small subset of the Facebook usership that forwards almost everything they receive to everyone they know. Pandering to that particular crowd is a Facebook developer's foremost goal, because they are the ones who will drive exponential growth, if it's going to happen at all.

I think market research will show that this core group of irritating people are just as capricious with the "block app" button as with the "forward". So assuming your idea is good enough to spread among the primary group, your first chance is generally the only one you get.

Re:

It's a small subset of the Facebook usership that forwards almost everything they receive to everyone they know. Pandering to that particular crowd is a Facebook developer's foremost goal, because they are the ones who will drive exponential growth, if it's going to happen at all.

So basically Facebook selects for applications that are attractive to the kind of people who forward spam.

Thanks for the warning.

most successful sites

can trace their success to that one weekend or month or season where things really took off

additionally, you misunderstand that the flash crowd is not something that comes and goes, but something that comes and stays

but sure, you are correct: a good innovative idea will find a way regardless of inability to scale quickly. some other guy will make work what you can't. you could retard your growth for awhile while you tinker with how to scale. but if some other guy takes your good, innovative idea and runs with it further and faster than you do, you are doomed to obscurity while he reaps the benefits of your good idea

so you shouldn't be giving advice on how the market works, because the fickleness you dismiss really is a big deal and is not to be taken lightly

Brand protection:

FTA:

including one that was (literally) Slashdotted

Anybody here think slashdot should be protecting it's brand here? Isn't this similar to using google as a verb? I think this is the only place one should be allowed to use that term. Microsoft most definitely shouldn't be allowed.

Make your application... efficient?

Then you don't need as much brute force ?

 

Great.

A paper on how to avoid slashdotting, posted on slashdot. /me clicks obsessively on links

The hard part is realizing that you need to scale

The problem with most sites is that they never expected the sudden burst in popularity, and having never bothered to test if their service was scalable, had to rush and fix it before people start noticing reliability problems. If you at least attempted to write a scalable app to start with, even if you don't have the details nailed down, you've taken a step in the right direction.

Animoto looks like an interesting case because it's a really resource-hungry app that has to put together a video with effects and music. Most sites have trouble just serving up dynamically-updated text. All those EC2 instances and the high-bandwidth needed sounds like a lot of money. Scaling up a business plan is at least as difficult as scaling software.

And for those of us who speak English?

Could someone provide a translation of the summary for those of us who speak English rather than promotional BS? .. on second thoughts, never mind.

or download the pdf

http://www.usenix.org/events/usenix08/tech/full_papers/elson/elson.pdf [usenix.org]

Re:Astro Turf

Doesn't Microsoft employ "bloggers" to seed pro MS babble to Web sites like Slashdot? Just sayin'...

If you're going to troll, it might be a good idea RTFA beforehand so that you don't make a fool of yourself. Two examples:

- The web service is implemented in Python and currently deployed on two virtual machines at Amazon EC2.
- Like Asirra, we implemented Inkblot in Python.

If they're astroturfing they aren't very good at it.

The article has very little Microsoft-specific details in it. It's basically a short explanation of high-performance content delivery and a few stories about MS Research [microsoft.com] (link because they have some cool stuff) projects and how they fared with high load traffic surging (aka Slashdotting). They specifically mention getting Slashdotted several times, as well as surviving a DDoS.

Overall I thought it was an interesting article. I didn't realize Amazon's S3 service was so inexpensive or available to "budget" sites.

Re:

Doesn't Microsoft employ "bloggers" to seed pro MS babble to Web sites like Slashdot? Just sayin'...

If you're going to troll, it might be a good idea RTFA beforehand so that you don't make a fool of yourself. Two examples:

Erm, for the record they also cover applications created by MS Research, using MS technologies - e.g. one called MapCruncher, which runs/ran on IIS. See 5 Application Design and Flash Crowd Experiences [usenix.org]. Another examples include SQL Server, but the article's pretty much technology agnostic

Re:

Actually, in this case I really wonder if he manages to make a profit. I looked at the amazon cloud recently, and if you really have one CPU unit running there non-stop -for year, it is pretty expensive, around 700 dollars I believe. For something like that you could also just get a dedicated rack server. So if this guy apparently needs a lot more than that, he sure needs to make quite a lot of money from his facebook app, because he will be spending several thousands of dollars on CPU power.

Re:

Hmm, you're right. Better revamp the method:

1) Be myspace or facebook
2) ride on the coattails of self-absorbed attention whores
3) ???
4) Profit!


5) throw doggie bones to javascript noobs so that sites' users could be spammed with even more movie quizzes.(You too can be a myspace developer! Valid e-mail address and fake phone number required)

Re:The method:

Well, you have to compare apples to apples with hosting, though. With Amazon's cheapest EC2 instance, you are looking at $72/month cost, or about $850/year. You have a bit more for storage, static IP, etc. But, it sure as shit beats the heck out of other boxes you could get at guys like GoDaddy and 1and1, where you get a shared box, with minimal control and cpanel or something at best, unlike root access at Amazon or admin access at GoGrid's windows boxes. You could go to services like Linode and get boxes with root access, but when you do the math, you will be hard pressed to get a comparable box at this rate. That, coupled with the flexibility of a pay-as-you-go model really does make this ideal for several situations. With the recent addition of persistent storage, you can even run full-blown db-driven apps here (something that was a pain in the ass before PS, because you had to use s3 as your permanent store). If you go to guys like Rackspace or other reputable providers, you are looking at $600-700 a MONTH as a start. Should you choose to own the iron yourself, you can probably get comparable numbers up to a certain threshold, but then you are left with hardware management issues. Of course, not that Amazon ever goes down ;)

Re:

hmmhmm, no no. I have a virtual server with root access and even ability to use serial access over ssh if everything else fails. This is about 9 euro per month. for 72 euro per month I can get a nice single core machine with 2 GB ram and 2X250 GB in raid 1. And I'm not even at the cheapest provider. Thing is of course, that such providers work with yearly contracts and you cannot catch peak usage as nicely as with amazon's cloud. I am just saying that when you as a garage shop get a large amount of traffic

Re:The method:

if you really have one CPU unit running there non-stop -for year, it is pretty expensive, around 700 dollars I believe.

That's pretty damn cheap. A dedicated rack server is upwards of $300/month most places, and it does not provide the "elastic" part of the Amazon cloud for when your service takes on heavy demand. Rackspace, for example, provides a comparable unit at $383/mo.

You might be talking about a Virtual Private Server--there are a number of services offering similar specs in the $120-200 range...still more expensive, but more comparable to EC2.

Re:

Softlayer [softlayer.com] has multi-core boxes starting at $150/month; we've got a box with them with a 15k RPM SCSI drive for about $300/mo.

For dinky personal projects, I've got a dedicated Athlon XP 2400+ with half a gig of ram with a little no-name provider [ezzi.net] -- and it only runs me $50/mo.

I've seen all sorts of prices in the $50 - $300 range for varying hardware. If you're willing to gamble on a lesser known host, you can get hardware cheap.

I wouldn't necessarily recommend running an established webapp with thousands of