Learning to scale.

My friend Christopher shared this on Facebook. It’s a queue for CXC results.

Tomfoolery. That’s the adjective that bubbled into my mind when I saw this screenshot and understood what it was describing.

Admittedly, this was a bit harsh. I mean, someone built this, they took the time to craft a way to deal with the rush of users trying to get their examination results from the Caribbean eXamination Council (CXC). They (CXC) have a classic seasonal scale problem.

During the year, the number of users hitting their servers is probably at a manageable level. And when results come up, they probably literally see a 10-fold increase of access request. They’d have the numbers, but I think this guess is reasonable.

I was curious about the traffic data on cxc.org, so I got this from similarweb.com

Their solution may have been reasonable in 2012, when maybe scaling on the web was more of a step-wise move in terms of resource management than a flexible thing as it is today. But there are solutions today that don’t necessarily mean an whole new IT investment.

I really dig this chart from Forbes demonstrating the seasonality challenge

So, how do you contend with seasonal IT resource demand without breaking the bank? Design your solution to leverage “The Cloud” when you need that extra burst.

“The Cloud” in quotes because if things were as easy as writing the line, “leverage The Cloud” then we wouldn’t be this deep on a blog post about it. So, specifics, here’s what I mean:

Plan the resources needed. In this case, it might be a solution that uses load balancing where some resource is handled on-prem and capacity from the cloud is used as needed. Alternatively, an whole new approach to sharing the resources at play might be worth investigating – keeping authentication local, but sending users to access results stored in the cloud via a serverless solution is a great consideration.

I don’t want to spec out the entire solution in two paragraphs. I do want to say CXC can and should do better, since they serve the needs of much of the Caribbean and beyond.

Advertisements

One Marvelous Scene – Tony & Jarvis

My video editing skills are what you might call, “Hello World” level. So, I’m not interested in bringing down the average as it were, in terms of the quality content over on the official “One Marvelous Scene” playlist on YouTube.

I dig what those storytellers did a lot. But as far as I can tell, even though there are 67 videos on that list, they missed a key one. It was from when Tony had returned home in Iron Man 1 and was building out the new suit, working with Jarvis.

Back when IM1 came out, I remember being delighted by the movie, just for the flow of it, just as a young developer then, seeing a portrayal of a “good programmer” was refreshing. And he was getting things wrong. Making measuring mistakes or pushing things out too quickly. He had to return to the drawing board a lot. It was messy. Just like writing code.

And while doing this, he was working along with Jarvis. An AI helper with “personality”. One that he could talk to in a normal way and move his work along.

In 2008, that was still viewed as fanciful, in terms of a developer experience, but even then I was somewhat seriously considering its potential, which is why I even asked about voice commands for VS on StackOverflow 👇🏾

Jump forward to 2019 and bot framework is a thing, the StackOverflow bot is also a thing and Team Visual Studio introduced IntelliCode as this year’s build conference.

So, as a scene in a movie, Tony talking with Jarvis helps us understand Tony’s own creative process, but for me, it gave glimpses into the near future for how I could be building applications. And that’s marvelous.

A short note on value-based pricing

This is a short note because I’m not an economist and don’t pretend to be one. I think areas like pricing and value and cost determination are complex and should be given their just consideration, however, I recently saw a question on Caribbean Developers and I wanted to share what insights I have on it.

When I first saw this, I knew immediately I wanted to say, “Aye Roger, lean away from thinking hourly rates”, but that felt a bit too curt.

It’s been my experience that freelance developers tend to think in terms of charging hourly rates and doing work that they cost based on that rate. The “better” or more experienced they get, the more the rate reflects their growth/maturity in the space. Then, they learn about markups, based on their understanding of customer risk and that good stuff. But I’ve come to appreciate that thinking in terms of hourly rates for the work you do in freelance is a trap. The actual term for what I think is a better approach is value-based pricing.

I found a great explanation of it on quora:

Value-Based Pricing means presenting a price to the purchaser which is based on their perception of the value they will derive from the result being discussed

More from David Winch, Pricing Coach on quora.

So, I think Roger needs to spend time understanding how to develop a firm that uses Value-based pricing. It’ll help much more in the long run.

After Roger posted his question, I saw a great point from @sehelburt on the matter too:

More developers need to understand Charge-ByValue not -ByHour https://t.co/ejO5hlu3l9

— Irwin Williams (@iStarr) April 26, 2019

Jumping over hurdles to get to insights using Azure

On the way to getting some data to an Azure DB, I explored strategy in using entity framework that was such a cool timesaver, I thought I’d share it for #GlobalAzureBootcamp

I could have done this years ago. It’s just that sometimes a task feels so complex, daunting or mind-numbingly boring that you make do with alternatives until you just have to bite the bullet.

We work in SMPP at Teleios. It’s one of the ways you can connect to mobile carriers and exchange messages.  From time to time, we need to analyze the SMPP traffic we’re sending/receiving and typically, we use WireShark for this. That’s as easy as setting it up to listen on a port and then filtering for the protocol we care about. Instead of actively monitoring with WireShark, we may at times use dumpcap to get chunks of files with network traffic for analysis later on.

What we found was that analyzing file by file was tedious, we wanted to analyze a collection of files at once. We’d known of cloud-based capture analysis solutions for a while, but they tended to focus on TCP and maybe HTTP. Our solution needed to be SMPP-based. So, we decided to build a rudimentary SMPP-based database to help with packet analysis.

That’s where the mind-numbingly boring work came in. In the SMPP 3.4 spec, a given SMPP PDU can contain 150 fields. The need for analysis en masse was so great that we had to construct that database. But this is where the ease of using modern tools jumped in.

I got a list of the SMPP fields from the WireShark site.  In the past, I would have then gone about crafting the SQL that represented those fields as columns in a table. But now, in the age of ORM, I made a class in C#. And if you’re following along from the top, I created a project in Visual Studio, turned on Entity Framework migrations and added a DataContext. From there, it was plain sailing, as I just needed to introduce my new class to my DataContext and push that migration to my db.

It probably took me 30 minutes to go from the list of 150 fields on WireShark to being able to  create the database without the necessary tables. Now, where does Azure come into all of this?

Each capture file we collect could contain thousands of messages. So, in my local development environment, when I first tried to ingest all those to my database, my progress bar told me I’d be waiting a few days.  With Azure, I rapidly provisioned a database and had it’s structure in place with this gem:

Database.Migrate();

That is, from the Entity Framework DataContext class, I called the Database.Migrate() method and any newly set up db would have my latest SMPP table. From there, I provisioned my first 400 DTU Azure SQL database and saw my progress bar for ingestion drop from days to hours.

With a database that size, queries over thousands of rows went by reasonably fast and gave us confidence that we’re on the right path.

We’re also working on a pipeline that automates ingestion of new capture files, very similar to what I did last year my Azure Container Instances project.

So, for #GlobalAzureBootcamp 2019, I’m glad that we were able to step over the hurdle of tedium into richer insights in our message processing.

Making a map.

I like how WordPress’ default on these blog posts is “tell your story”. LOL.

Recently, I described wanting to use Azure App Insights for dependency tracking. I started by looking at documentation on it. I ended up using this, along with this video.

As I started looking into the first place to drop in dependency notification, which was where I create my Azure Container Instance. The tutorial’s code suggested dropping this fragment into a finally around where the dependency was, like so:

From: https://docs.microsoft.com/en-us/azure/azure-monitor/app/api-custom-events-metrics#trackdependency

Then I found myself wondering if for the dependencies I was going to track, wouldn’t a C# attribute be really good. Then I fell down a rabbit hole of when to create attributes.

And then I emerged out of that hole, attribute-less.

So, after I applied guidance from the Application Insights document, I colored my external operations with track dependency calls and ended up with this image:

Azure Application Insights application map for my solution

This does what I wanted, which is to visualize the components and dependencies of my solution, with information on success/failure and duration of the calls.

I’d like to do more with this.

Conquering complexity with a map

Last year, I worked with a researcher to develop a really cool, complex Azure solution to augment her work flow with some data. She needed to ingest a large volume of data, clean it up, run some AI on that and then present it. Typical data science activities that she wanted to run in the cloud.

I implemented the flow using several components including Azure Container Instances, Event Grid, Azure ML, Cosmos DB and Azure Functions. My daily drive at work doesn’t necessarily let me play in all those spaces at once, so I felt really glad to see all of those pieces work together.

Deploying took a bit more work as I wanted to make that as straightforward as possible. Thus, I used the Azure Fluent SDK that I was fanboying about across a few posts in 2018.

After everything was deployed though, I found visibility into the system was a bit of a stretch. My researcher colleague wanted to easily know where things were at in a given iteration of the process. I didn’t have a lot of solutions for that, so it was mostly email alerts.

That is, until I learnt about Azure Application map from two of my colleagues at Teleios – Ikechi, in Ops and Anand in Engineering.

It’s a part of Application Insights and lets you track dependencies between various services in an Azure solution. So, just out of the box, you can view the success of calls between web sites and web services and databases and other components. Going further, you can even add other components and dependencies to your app. That got me thinking. Maybe I can use Azure Application Map to display the various components of the solution and track issues in a visual, at-a-glance way?

I’m going to check it out.

Project VILMA

Sometimes, I get home and just wish I could say to my virtual machine in the cloud, “magico presto, turn on!” and it’s on and ready for me to remote into and do things. I wanted to do things to make that happen, but time and procrastination happens to many. Thankfully, there was an upcoming developer gathering that I used as the catalyst to actually build a system that would work, almost like magic.

So, last Sunday, the Port of Spain chapter of GDG (Google Developer Groups) held a developer event, #GDGPOS #DevFest. They reached out to the local developer community for interesting projects and I responded with a proposal to build something that would work in the way I described.

GDGPOS Presenters

My proposal got accepted and I spent a few weeks building out the idea. My whole solution involved using my Google mini to turn my virtual machine on or off.

To do that, I created a Google Action on the Google Actions console. I had played around with Actions before, but this would be different. I have been making most of my conversational agents using Microsoft’s Bot FrameworkBot Framework, so a lot of the concepts were familiar to me, from things like Intents, to Utterances and even the use of webhooks. For this action, I largely had to focus on just one Intent – the one that would hear a command for a VM state change and execute. Overall, the system would look like this:

• Creating the action

So, I created this custom Intent that took me to Dialogflow, Google’s interactive tool for building conversational interfaces. There, I created a custom intent, ChangeVMState.

ChangeVMState would receive messages and figure out if to turn a VM on or off. The messages could be in a range of formats like:

• turn on/off
• power on/off
• shutdown/start up the vm

They would resolve to the ChangeVMState intent. All messages sent to ChangeVMState was then forwarded to my webhook. I deployed the webhook as a function in Azure.

The code to execute the functions is pretty straightforward. One function receives the request and queues it on an Azure Storage Queue.  Azure functions provides a really simple infrastructure for doing just that. 

I mean, this is the whole method: 

The item being put on the queue – the desired VM state – is just a variable being set. 

Another function in Azure will then take up the values in the queue and will start or stop the VM based on state. Again, a pretty simple bit of code. 

I’m using the Azure Fluent Management SDK to start/stop a VM

So, finally, after the VM is put into the desired state, an email is sent either saying the VM is off or that it’s on and an RDP file is included. Ideally, I wanted to have the Google Assistant I was using notify me when the VM got up and running, but I just couldn’t get push notifications working – which is why I ended up with email. 

Thus,  I ended up with a Google Action, that I eventually called VILMA Agent (at first, I was calling it Shelly-Ann).  I could say to my Google Mini, “OK, Google, tell VILMA Agent, turn on” and I’d get an email with an RDP file.

The code for the functions part of VILMA is up here on GitHub.