Disqus - Latest Comments for shalunov

Re: Three paths of B2C startups growth: ads, free ads, and viral

shalunov — Mon, 13 Jul 2009 03:47:14 -0000

Spam.

Re: Mean Delay Considered Harmful

shalunov — Mon, 06 Apr 2009 08:22:31 -0000

The network with an effective 40% loss rate is quite unusable. The network with an effective (round-trip) 5% loss rate is, perhaps somewhat surprisingly, still effectively unusable for any interactive app.

Consider an HTTP request that gets 20kB of data. Without loss and without cached ssthresh, this will take five RTTs not counting the FIN exchange. (One for SYN/SYNACK, and then four for the 14 data packets, with slow start doubling the window at every RTT and sending an incomplete last window.)

Consider a web page. A very minimalistic one might take 25 RTTs (include some DNS and a picture or two and maybe a stylesheet). With objects this short, fast retransmit can't work. (HTTP/1.1 running over TCP with real SACK support could help a bit here, but let's get back to this universe.) Assuming uncorrelated loss, that's 72% probability of going into timeouts, which will dominate the total user-experienced delay.

You want 99th or 99.9th percentile for what you're trying to do, unless the application is very loss-tolerant, like certain VoIP codecs which can take about 5% and still work OK.

Regardless, the choice of percentile to report should not be driven by the desire to encode loss into it. Nor should general-purpose metrics depend on the app too much, and loss tolerance, because it is easy to dial in, varies greatly.

The median is at least more robust than the 95th percentile.

The better place to have this discussion is the IPPM WG of the IETF where http://tools.ietf.org/html/draft-ietf-ippm-repo... is close to last call.

(Incidentally, the IPPM Surveyor, one of the first Internet measurement systems to report percentiles, used 0th, 50th, and 95th in its plots.)

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

michael — Sun, 15 Feb 2009 00:50:52 -0000

wisconsin is one of the best mathematics schools in the world, and moscow state university is one of the best universities in the world.

Re: Choosing a web dev platform: Rails or PHP? ԅ Hacking Startups

shalunov — Sun, 18 Jan 2009 02:08:30 -0000

PHP is easier to deploy, you're right.

By change I meant substantial requirements change. Suppose you build a blog platform, then decide you actually want to add a social network on top. Migrations in Rails make this easier.

Does this make sense?

Re: Choosing a web dev platform: Rails or PHP? ԅ Hacking Startups

shalunov — Fri, 16 Jan 2009 21:18:07 -0000

Nivi, thank you. Because free ideas and reasoning are more useful than free advice, I try to provide reasoning. As a side effect, it helps articulate the assumptions.

Re: Choosing a web dev platform: Rails or PHP? ԅ Hacking Startups

shalunov — Fri, 16 Jan 2009 21:07:22 -0000

Perl has lost mindshare. Five years ago, you'd bet any CPAN module would at least *compile*. When I last moved a Perl app, that was no longer the case. We later rewrote the app and stayed with Perl (I argued for staying then, and still believe it was the right choice).

I didn't recommend Cake, Django, or Catalyst as options. Rails is the original and just as free. Googlefight [ruby rails] and [perl catalyst] for a rough measure of mindshare.

Re: Choosing a web dev platform: Rails or PHP? ԅ Hacking Startups

shalunov — Fri, 16 Jan 2009 20:56:09 -0000

Rails isn't just a framework. It's the killer app for Ruby, and it got cloned into every other web dev language.

You can DYI if you care about making tools, but you shouldn't if you care about making apps.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Sat, 10 Jan 2009 00:00:56 -0000

My sample size is well over a million of users of my apps, which range from very serious to lightweight to utterly giggly. Yours is two dozen former classmates.

Giggly users are by far more viral.

The post was meant for fellow developers and to Facebook. For what it's worth, developers who made apps used by vast majority of Facebook users agree with me, but Facebook obviously didn't. At this point, it's history.

P.S. I went to the best school there was in Soviet Union, which was not open for foreign travel. My stint in Wisconsin grad school was boring -- I passed the PhD quals before the first semester started. At least I did math, not acoustics and baseball.

Re: Mean Delay Considered Harmful

shalunov — Fri, 09 Jan 2009 23:06:14 -0000

Look at the difference in phrasing. It's not accidental.

"Has a heavy tail" is a precise statement about the hypothetical underlying distribution from which the samples are drawn -- that this distribution has an infinite deviation.

"Tail is relatively heavy" is a quantitative statement about the empirical distribution -- that a good chunk of the total delay comes from a few samples. This sure looks to be the case in the data that I've seen.

The mean can have its advantages -- obviously mean(x+y) = mean(x) + mean(y) is an important property, with nothing corresponding for the median, which needs a convolution here. So I don't say that median it universally better, only that it's better for network delay. Similar considerations apply to many other sample sources, of course.

Re: Net Neutrality: Three Questions

Robert J Berger — Mon, 05 Jan 2009 03:24:20 -0000

Stanislav is correct. A fully capitalized fiber to the home can be done for $3000 fully loaded. The $3k can be amortized over 10 - 30 years as the fiber can have increasing capacity over time just by changing the electronics on the ends. Today to deliver 1Gbps over "last mile" distances (ie 3 - 8 miles) is pretty trivial.

It would be cheap (especially compared to the lunatic spending to bailout dead industries) to get fiber to most homes and businesses in the US. But not if said fiber is owned by the existing CableTelco oligopoly.

Re: Net Neutrality: Three Questions

shalunov — Tue, 04 Nov 2008 17:03:48 -0000

Gigabit Ethernet is the current cheapest last-mile technology for new installations and continues to be. For short distances (under 300 meters), one could even use multimode fiber. My friend and I prepared technical documentation of how to accomplish what I describe here for the Internet2 testimony at the US Senate net neutrality hearings. This withstood expert scrutiny during intense debate.

The relationship between the amount of the Internet traffic and the capacity of a single fiber strand is surprising, but no reason to be rude.

One strand of fiber can carry, with DWDM technology that became commercially available at the turn of the century, 160 wavelengths at 10Gb/s each. Currently, one could do a few times better, for example, by lighting the wavelengths with 40Gb/s OC-768 instead of OC-192. This was still cutting edge when the article was written, and so I assumed OC-192. That's 1.6Tb/s. This exceeds Internet traffic in 2006 -- see Andrew Odlyzko's Minnesota Internet Traffic Studies, for example. (Today's traffic is higher, but because of OC-768s becoming routine production technology, it still should fit into a single strand.)

Re: America Needs A Turnaround Plan

shalunov — Tue, 07 Oct 2008 21:25:00 -0000

Russia has since then paid off the debt with the petrodollars. The US, as an oil importer, of course, is on the other end. But don't envy the petrostates -- none of them are in a particularly good shape even today with very high oil prices.

US sovereign default would only be possible through a combination of congressional and executive hiccup (because the Fed is "independent", both branches would need to forget the bill). The treasurys are denominated in US currency -- a situation unique for sovereign debt. The US can so far simply print more dollars and inflate its way out of debt.

Re: Ink Test

shalunov — Tue, 16 Sep 2008 17:42:12 -0000

150 days.

Re: As Facebook hits 100 million user mark, twenty percent have already opted in to the redesign

shalunov — Tue, 26 Aug 2008 02:24:23 -0000

The users will probably first hate it, complain about it, and then get used to it. The way I saw a user describe her new profile: "Why can't I add [app X] to my profile? It was there before."

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Sat, 05 Jul 2008 19:10:21 -0000

Check ages outside of the US -- say, UK or Canada. Much closer to general demographics.

Keywords that users are supposed to enter into their interests are garbage, which is why there is no premium in targeting for particular ones. If you have an investing app, fetch interests of its users and look at fraction who have entered "investing". Other than you, anyone?

There's clearly carrying capacity for serious apps. Different estimates range from pessimistic 5% of users ("just" a few million) to 30%. Giggly apps are more viral, but if they competed in separate categories, this would not matter.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Thu, 03 Jul 2008 23:24:41 -0000

Serious apps exist: http://blog.shlang.com/post/40911109/useful-fac...

There would be more if they could spread and developers would have a reason to make them.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

tomo — Thu, 03 Jul 2008 22:32:40 -0000

Great post. But does Facebook have any serious apps to group together? I'd be curious to see them. Maybe all the glitter has obscured from my view.

Also agree with Dave I think LinkedIn has already taken the mantle of serious users' socnet. I agree many people use LinkedIn for job hunting, but LinkedIn has a lot of services for corporate clients now, so LinkedIn doesn't have to depend on ad revenue the way that Facebook does. Because of the ad-based model, seems like FB is happy to go for the largest possible audience, which as you point out skews giggly not serious. Also in current form it's hard to see any company using Facebook for recruiting, business intelligence research, etc.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Thu, 03 Jul 2008 19:53:29 -0000

Yes, the giggly and serious parts of Facebook indeed resemble MySpace and LinkedIn. But look closer:

serious Facebook >? LinkedIn

Most people basically only use LinkedIn when looking for a job. Serious Facebook offers much more and gives you a reason to visit often. Currently, however, because of the giggly noise, serious Facebook does not live up to its promise of facilitating day-to-day networking and keeping in touch with people instead of job-hunting.

(Yes, LinkedIn cloned a bunch of Facebook features and this made LinkedIn more useful. My point exactly.)

giggly Facebook > MySpace

Giggly apps make Facebook fun. Infinite supply of quizzes, fun pokes, great mechanisms for forwarding pics.

Facebook could solve the ">?" problem for serious users without heavy-handed TOS police. And it would actually make Facebook *more* fun for the giggly users, because developer incentives would be better to create more glittery quizzical pokes for them.

I do agree that it's a branding challenge. But perhaps surmountable one. Perhaps make them into two modes and talk about how Facebook is both for work and for play.

But first things first for Facebook.

Step 1. Acknowledge that giggly users exist and comprise the majority.

Re: Would you work with micromanaging boss, no salary, and all your work thrown away? ԅ Hacking Startups

shalunov — Sun, 22 Jun 2008 17:27:58 -0000

Jen: Socialization, as in becoming socialist like the teacher, is, of course, harmful.

Having friends and peers who inspire you is very important for most people, because people tend to define their goals and references relative to peers. Which is why being able to read a book a week or do elementary algebra or program a simple game are supposed to be achievements for kids her age rather than matter of course. Randomly selected peer group is, of course, less desirable than one she can choose.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shiva — Sat, 21 Jun 2008 23:03:38 -0000

very true.. it is not just apps on facebook, but it is symptomatic of the web itself. It is the classical noise vs signal, and what is noise and signal changes based on user preferences. I think, this is something, that has to be built into the fabric of web browsing a.k.a browsers.

What we really need, one unique location to learn about current users preferences (read as openid, with preference/profile plugins, hosted on a unique url for every user). There is a product waiting to happen right there!

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Sat, 21 Jun 2008 01:30:27 -0000

The problem on Facebook is worse than on a typical web 2.0 site.

Facebook wants serious and useful apps, but they can't exist while squeezed out by more viral giggly apps.

Re: 75% of Facebook users are giggly and poke; 25% are serious and import bookmarks ԅ Hacking Startups

shalunov — Sat, 21 Jun 2008 01:27:23 -0000

Facebook already provides self-tagging tools (interests, favorite books, etc.). Just a touch over 1% of Facebook users understand this. Even many geeks can't figure out that commas in Facebook separate tags that you can search on by clicking.

Regardless, the number of disjoint categories needs to be minimized to maximize potential app reach, which is a developer incentive, and competition in each category.

Re: Would you work with micromanaging boss, no salary, and all your work thrown away? ԅ Hacking Startups

shalunov — Sat, 21 Jun 2008 01:17:41 -0000

My proposed solution to people stabbing themselves in the eye en masse is, of course, to stop stabbing.

Re: Would you work with micromanaging boss, no salary, and all your work thrown away? ԅ Hacking Startups

shalunov — Sat, 21 Jun 2008 01:13:13 -0000

The employee produces as much value as the boss (none), but the boss gets a small salary and a full benefits package.

Whether kids really aren't capable of doing anything useful is open for debate. In any case, aiming at doing useless crud is guaranteed not to produce any value. Might as well not bother.

Re: Private Pastes (and Projects) for codepad.org

shalunov — Tue, 17 Jun 2008 21:12:08 -0000

But now, what are you going to charge for?