Someone once told me: before you look forward, you should always take a look back...and make sure nothing is gonna bite you in the ass! (They also said you should spend most of your time looking through the windshield and not the review mirror, but hey...it's my blog post).
My take is that a bit of self reflection is especially valuable in software development and with startups in general, where technical debt and false assumptions about market traction can trip you up right when you need to grow fast.
Taking stock of your success is equally important, helping you benchmark velocity as you plan your next moves. Heck if we did that stuff with only 10 smart people, imagine what 20 properly resourced ones could pull off!
So as Packet looks forward to the coming year, we’re also looking back at what we accomplished - and what we didn’t - over the past twelve months! Hint: if you want to see our failures, skip straight to the bottom. ;)
What We Nailed, Month by Month
January - The Type 0
We released the Type 0, a $.05/hr bare metal server, which quickly sold out. Our “Tiny but Mighty” has since gone on to be a key part of our portfolio and a critical piece of our customer journey, serving as anything from a bootstrap node, a testing box, a load balancer, or a private OwnCloud server to an affordable network throughput monster.
February - PB&J and Tinkerbell
As part of a redux of our entire provisioning system, we released two internal microservices: PB&J (power and boot control) and Tinkerbell (our Golang take on a PXE server). These made server installs faster and more reliable - nothing sexy there, but perhaps the most awesome accomplishment of the year that nobody knows about.
March - Two New Datacenters, Plus Block Storage
We went from one datacenter to three, adding SJC1 and AMS1 in the same week. Our block storage service also jumped out of beta in EWR1.
April - Dual Launch Partnerships
We partnered up with StackpointCloud and CoreOS to help people deploy "Kubernetes in three clicks" via stackpoint.io. The bonus? We upgraded all of our servers except Type 0 to include a TPM chip, allowing Trusted Compute in CoreOS Tectonic. With Mesosphere, we helped whip up a Terraform manifest to enable easy deployment of DC/OS clusters.
May - New Provisioning System!
We ripped out the last of our legacy code and implemented a new image based installation service (Casper), Penseive (automagically creating rDNS records for each server), as well as our rescue mode microservice (S.O.S - serial over SSH).
June - Type 2’s
After much anticipation, we launched our first Dell based server, aimed at virtualization and private cloud workloads. The Type 2 has since become our best seller - we literally can’t keep them in stock.
July - SoftBank Funding
We were thrilled to find in SoftBank a strategic investor that had ambitions as big as ours. Along with our seed investors, SB helped us raise a total of ~$11MM to fund the next stage of our vision for automated fundamental infrastructure.
August - Vacation?!
Seriously, we can’t remember doing anything in August except for selling, hiring a new head of Engineering, doing tons of paperwork, and buying a lot of hardware. But that’s about it.
September - Windows on Demand & FreeBSD
We expanded our list of supported operating systems to include Windows Server 2012 and FreeBSD. Among all of the Minecrafters and HyperV aficionados, there was much rejoicing!
October - Spot Market for Bare Metal
Working with our friends at SpotInst, we released the first spot market for bare metal compute, allowing dynamic consumption of reduced (and variable) rate capacity.
November - ARMv8 Server
We pulled the cover off of our ARMv8 powered “Type 2A” - a 96 core beast that, well, nobody quite knew what to do with. Until we invited the Docker pirates, the gamers, and the big data folks, who started making things like Rust, Android, Flynn, and Docker work.
December - Tokyo Datacenter & Self Service BGP
We ended the year by opening our 4th datacenter (NRT1) in Tokyo, Japan and shipping our biggest networking feature yet: self service local BGP and support for global announcement of custom ASN and IP space.
December is also a time for parties. We had two of them! Our friends at SoftBank, Foxconn, Cavium, and CreationLine helped us throw an excellent Tokyo launch party just days after the NYC community came together with Packet and partners Aporeto, Tigera, NYNOG, Myriad Supply, Apprenda and CoreOS. Much love, much amaze!
Five Things We Whiffed On
As always, there are the things you look back on and go: ouch. Here’s what we didn’t get done, or didn’t do good enough (but we’re sure going to tackle in 2017!).
- Over Promised - We often thought we could ship features faster than we really could. Things like block storage, Type 2 servers, and image based installs at scale simply took longer to get right.
- Failures - We aim for 100% success on provisions, but at times we got as low as 95%, which made it hard to sleep at night. People need their bare metal!
- Storage - We heard from users throughout the year that a storage heavy server option would be valuable. The problem was (and remains): everybody has a different perfect storage server! We’re a bit later than we want, but we think our new Type S (arriving in early 2017) will suit a broad range of use cases and fill a gap in our portfolio.
- Network Insights & Visibility - We had hoped in 2016 to dive deep into our network and share with users metrics, traffic insights, and other valuable (tid)bits to help users unleash the full potential of our platform. Ends up we were too busy opening up datacenters and building the rest of our platform to make much progress here. We plan on it for 2017!
- Capacity - Growth hit us hard this year, and the realities of financing and shipping around the world millions of dollars of gear resulted in a few periods of “out of capacity” issues. With tools like our new spot market, we hope to even this out in the coming months despite rapid growth.
How was your year? What did you win and fail at? We'd love to know!