Friday, July 31, 2009

How Fooala SCRUMs

In this post, I would like to touch on how we use SCRUM to manage our development @Fooala. And expound on the lessons I have learned as the SCRUM master. (If you are not familiar with SCRUM check out this video. Also, here's a useful diagram.)

(Note: while SCRUM master, I maintain a development role as well. I am liaison between business development and software development making sure both align.)

How do we SCRUM?
Our daily SCRUM occurs at 2:15PM.

Sprints are 4-7 days long. Stabilization sprints are 1-3 days.

Releases every 2-3 weeks.

We implemented SCRUM without any software. We use a whiteboard and post-its.

Post-its define tasks. Each task is placed on the white board and belongs to a developer. As the task progresses, it moves states as follows: "Backlog", "In Progress", "To Verify", and "Done!"
We write the expected amount of hours each task will take on each post-it and modify them every day during the daily SCRUM.

Different colored post-its are used for each of our products.

We also maintain another whiteboard that keeps the task backlogs for the next two sprints.

I have stapled legal paper around our whiteboards in order to group tasks together.

Finally, we maintain a sprint burndown chart in excel and print charts every day or two.

Why keep SCRUM physical?
By keeping SCRUM physical, the tasks feel more tangible and have weight. It is also easier to visualize our progress. Often times, I catch another cofounder working on development turn around from his desk only to stare pensively at the SCRUM board. For him, seeing the tasks move from left to right is a physical movement that parallels our development progress.

Adding tasks to the SCRUM board requires physical effort so tasks are not added without some consideration. This helps to keep tasks focused, relevant, and manageable. However if we are away from the office, SCRUM tasks are emailed to me but I enforce that task be defined only in the subject line of the email.

Why SCRUM at 2:15PM?
The afternoon daily SCRUM is largely due in part to our developer's vampire-like work schedule. As a manager, I would love to have the daily SCRUM at 10:15AM and keep core hours 10AM-4PM. However, fighting the developers would be futile. Core hours are Noon to 6PM. Typically, the developers are up till 3AM, get to the office by Noon, have lunch, work till 6PM, have dinner and relax, then start working again around 10PM. SOOO, I have no choice but to work around the developers schedule. Having it at 2:15 makes sure the developers are awake.

Tasks vs. Features
Sometimes a task is added that is expected to take 20+ hours. When a task is expected to take 20+ hours, it is more likely a feature. Hence, a feature should be broken down into several tasks. When I realize a task is really a feature, I ask my developers break down the feature. If they can't do it, that usually means a design meeting needs to be held.

The dreaded task pushback
At times, it is necessary to take a task out of the current sprint and move it back to the next sprint/release. It is tough to know when to do this, but this is where the burndown chart helps.

The burndown chart illustrates how much effort is left. If i realize that physically there is not enough time left in the sprint to accomplish that much effort, then I know something's gotta give. Largely based on business goals, I move tasks back or mandate certain ones be given priority. And if need be I proclaim that tonight is going to be a late night at the office meaning dinner at the office, coding party till 5-6AM.

Releases
Planning a release starts by first evaluating how the last release went. We do a release post-mortem answering questions like: 'What went well this release? What can we do better?'

Then we start with what are the business goals we need to accomplish. We then select the features that align with our business goals. Finally we breakdown the features into tasks. The tasks can be pre-existing (either pushed back or on one of the legal pieces of paper). However if a new feature is being developed, a design meeting is scheduled.

We end our releases with a Release Demo. I try to add weight to the Release Demos by inviting our advisors. Usually a business/sales person produces a slide deck and gives the presentation. My belief is that if our business people don't understand what the developers have done and the impact of their work, then the overall vision of the company cannot be aligned. Also, by emphasizing the importance of the Release Demo and including our advisors helps motivates everyone both developers and business people.

Sprints
Like releases, sprints start by evaluating how the last sprint went. Mostly I see what has been pushed back from the previous sprint and whether it is still important.

Sprints end with a demo day that includes only the developers. Each developer spends 10 minutes demo-ing their work. (Applause is encouraged)

Stabilization Sprints
Finally, at the end of the release, we have a stabilization sprint. This sprint is meant as a buffer to help clean up code before it is rolled out to production. Admittedly, this sprint should not exist and should be taken care of while a task is being verified. However, I find it is necessary to keep up the extreme pace of development.












Thursday, April 2, 2009

Distributed System for Photo Mosaic Processing

My passion for photo mosaics and highly scalable distributed systems has led me to ponder about a distributed system for photo mosaics. Here's a first go of how it would work.

Assume the photos reside in some warehouse available via a REST interface. Assume also that all the photo indices can be received in one get request.

For now, we will assume a star topology with a master node that distributes all work to several worker nodes.

Abstract
A collection of photos is defined. The photos are distributed to several workers. The workers measure the "colorness" of each image and store the colorness with each image locally. The images are then sorted locally and a parallel sorting algorithm is used to sort images globally. Each worker receives subsections of the global image and determines the "colorness" of each subsection. The subsections are then sorted locally and a parallel sorting algorithm is used to sort subsections globally. The subsections and the images and then paired based on their order after sorting. Finally the images are assembled based on their paired subsections original global image placement.





Saturday, March 28, 2009

There is no formula

I recently watched the documentary Before the Music Dies. The documentary had many insightful comments about the record industry. In particular it focused on how the quarter-by-quarter profit mongering corporations sucked the life from music by caring only about the bottom-line.

With companies like clear channel focusing more on playing music that the masses "won't dislike rather than music the masses might like", music has become stagnate, at least popular music. But welcome the internet and music post-Napster.

Musicians now can produce a high quality album with only a few thousand dollars. They can market their music on myspace and youtube, spread their fame through Facebook, and reap profits on iTunes. This method works well for smaller artists whose music cannot be found on torrents for the most part. And even for the big names whose music is spread across the torrent cloud, they make reasonable profits from legitimate sales on iTunes or rhapsody or amazon.

I'm curious if record labels have decided to raise concert ticket prices to make up for the income they have lost from selling multi-platinum records. It would make sense to try although it might just annoy consumers who will relive their live experiences on youtube.

This documentary shed some light on what's broken with the music industry, or rather what has changed. Record labels can no longer force consumers to listen to what they deem is cool in order to sell more pepsi products. There now exists a meritocracy within music where only the most talented, or by today's terms, most viral musicians survive, e.g., 'Chocolate Rain'. But 'Chocolate Rain' is crap as is most viral music which is why we still need record labels to help consumers filter. But I argue that this is all a record label should do. Filter for talented artists and potentially great music and provide for them a nurturing environment rather than exploiting musicians for the sake of selling an ad.

On the other end, musicians need to write music that matters to them, that they connect with and not worry about catering to the masses in the hopes of making it rich. If by writing for themselves, they are able to connect with 1 or 1 million other persons then they have achieved their goal. Music comes form their heart, as in many of life's passions. You can not calculate it. There is no formula.




Thursday, March 26, 2009

LaTeX + Evernote = better math notes

Evernote is a great way to push your notes to the cloud. And the content you create is indexed for later searching. LaTeX is the standard document markup language used in scientific research papers, or any other document that requires some math. However taking notes in LaTeX is not really feasible since it is not very easy to type documents on the fly. Even editors that provide buttons to insert symbols are hard to use and slow down note taking speed. The ideal solution is to simply be able to write notes like with a tablet. However the notes you take should use some sort of recognition to convert the notes into LaTex. From there, the documents should be uploaded to evernote and indexed.

Hulu + Tokbox

I thought of the idea of being able to watch Hulu and tokbox at the same time. The key idea is that both people can control the video just like you already can with youtube. I'm not sure how to do this though. A simple idea is that at the least a web application can be created where hulu videos are embedded and tokbox parties are also embedded. That's the simplest mashup I can think of. But both people will have to start their video separately.

The way I think tokbox is able to synchronize youtube watching is that they buffer it on a separate server first. This is based on the different control one sees/uses when watching within a Tokbox party. So perhaps a more functional solution would be to have a server that routes the traffic to those in the party. Still not sure how to handle the video routing.

iPhone Developer Tips

Out of curiosity and the desire to expand my resume, I am trying to develop an iphone application. After watching some tutorial videos, I was able to open dashboard and run a simple application with a basic pane. Before getting knee deep in objective-c code I've decided to take a lesson from my past experiences of learning new programming languages. Rather than simply going straight at the code, I want to learn all the quirks of the language first. I have found iphone developer tips. Hopefully after reading a few of these tips, developing an application will be less difficult. My first application, a simple web app based control system for my ECE135 project.

Wednesday, October 15, 2008

Pervasive Computing 2.0

Currently, I am taking a course titled Emerging Mobile and Distributed Systems. The class focuses on pervasive or ubiquitous computing. The fields founder, Mark Weiser, once described how "the most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it." Weiser's very apt and concise comments have inspired several engineers but I highly doubt ubiquitous computing will reach the level he conceived. The problem? Social ignorance and selfishness.

Let me first describe the notion of social selfishness. Individuals act greedily, acting upon impulses with foreseeable, immediate benefit. Systems where users provide information that other users can then use later, such as
Cooltown, will ultimately fail unless each provider gains something. What will work is a system where users gain something immediately, i.e., entertainment, money, a relationship, or perhaps a service. If a user is to post something for the benefit of others he must find at the very least some entertainment value. He should consider his work as fun and therefore his time spent worthwhile.

My second argument is social ignorance. Let me briefly describe what I mean by it. Concerns over privacy, client thickness, energy management, etc... do not really matter as long as what's available is the social norm. For example if a battery only lasts 10 minutes when running GPS, Satellite TV, and
XM radio, but every device on the market has the same limitation, then users will not care. A more important example, is privacy. If every location based application requires full disclosure of location, then no user will worry about sacrificing his privacy. But, the applications must provide some value or else why sacrifice privacy at all in the first place.

To understand how these two social constructs will hinder ubiquitous computing, let me argue against the
CarTel system. CarTel is a system where by users with GPS tracked cars relay their position through open WAPs. The system works great if enough cars have WIFI enabled GPS but most don't. And social ignorance will make users not pay more for WIFI GPS unless either it's standard in every GPS device or they reap more benefit than simply knowing traffic patterns (social selfishness). Arguably one could use a GPS enabled smart phones but proliferation might take a while.

Another example is LBS social networking. Users willing to participate in a LBS social network like
Loopt receive benefits of knowing which friends are nearby, ratings of nearby restaurants, taxi hiring, ridesharing, and neighborhood announcements. Arguably each of these applications can be designed separately except the nearby friends feature. I argue this feature is not enough to persuade millions to use Loopt. Having friends or family know exactly where one is at all times is not necessary and has more risk than reward. Loopt mix however that allows users to meet new users with similar interests could work if meeting someone online had less stigma. However, if meeting someone online was the norm then social ignorance would precede past judgment and users would use the service.

Another example is the omniscient pervasive computer much like HAL 9000 from 2001: A Space Odyssey. HAL would never work. Mostly because humans cannot design a system smarter than a human. HAL would have to differentiate human emotion, something humans have difficulty deciphering. Furthermore, unless there was a standardization across all platforms of technology, HAL would not have the contextual awareness or balanced
proactivity to help users without annoying them. Again HAL must act smarter than humans. A simple version of HAL might be the robot in the fifth element that automatically cleans up broken glass. To create such a system would require either excessive technology or extensive training neither of which would coincide well with social selfishness.

What will work in ubiquitous computing? Arguably no application that tries to connect a room full of strangers will ever work. Society will never allow random individuals to meet random individuals. It's the same reason we don't introduce ourselves to every living being we can see. Hence social network used as a filter not a connector is what will matter. Secondly notions of publicly shared hardware will never work again because of social selfishness. Hence what
nees to happen is the growth and development of mobile devices. We are already seeing this with Android and iPhone. However application development should be cross-platform and until we see a standardization on platforms, perhaps even just a JVM, social applications will not reach critical mass. Only social applications with enough mutual benefit will work. And ubiquitous systems that try to connect users to the physical environment must not only be seamless as Weiser wisely describes, but have a significant benefit.

In conclusion, a large part of Ubiquitous computing tries to design a system that allows people to either be more time efficient or to allow effective use of slivers of time. This notion seems to be
biased to American workaholics . Would a lackadaisical European ever bother during his summer to worry about being connected? Ultimately the most pervasive system is Mother Nature. Not until embedded systems make me wonder my existence or impress upon me a ray of warm sunshine will I agree that ubiquitous computing has succeeded.