The HN Effect and Our Accidental Launch
Mostly due to the fact that neither Anton nor I are cut out to be movie directors, producers or presenters, we submitted our YC application in the last moment at five minutes to 8pm PST. This was 6am in Moscow on Monday.
Having only slept for 6 hours in the preceding 48, I hit the bunk right away, only to be awoken a few hours later. I stumbled to the computer to check our Clicky logs to see if anybody from YC had already looked at the site. To my surprise, we had 200 visitors - twice the average daily uniques.
At this point it became apparent that @yogstoth had submitted us to HN. I should say resubmitted, because Akshell has been on HN a few times before. In fact, Anton and I "met" in the comments of the first submission.
48 hours later, we have seen 20K+ visitors, with 300+ concurrent users online at peak, average time on site at 4+ minutes, a few investors knocking on the door and bloggers asking for interviews.
What Went Wrong
- we had no alerts in place to notify us of the swarm of HN and twitter visitors - I was asleep and Anton was on the subway when things took off
- we had no monitoring other than top and Munin which could only be accessed from localhost so was as good as useless as I couldn't set up an SSH SOCKS proxy
- we didn't have the necessary infrastructure nor a clear step by step plan for how to scale without downtime - the service was running on the cheapest Linode available
- we didn't have a mechanism for capture user info other than getting them to register after creating a webapp; as a result our conversion rate from visitor to registered developer was below 1%
- we didn't make it easy for people to Like or RT our URL, despite having this feature on the TODO list
What Went Right
- we responded quickly by replying to people, posting the HN comments link on Twitter and continued engaging users on all fronts
- our landing page was simple and to the point; we made the barriers to entry as low as possible by letting users write code and run the webapp without registering
- people actually liked the service
- we coped with the load by: quickly adding cloudkick to improve our monitoring, asking for advice on #linode on OTC and consulting Stephen a #devops expert from Atalanta Systems, drinking a lot of coffee and not sleeping for another 48 hours
What We Learned
- you should reply to people with questions
- you should make it as easy as possible for people to promote you
- you need both Cloudkick and PagerDuty set up before you have an emergency
- you should find the fastest solution to your most pressing problem: we considered upgrading our Linode instance which would've resulted in 15 minutes of downtime, but instead opted for a RAM upgrade which was a 30s or so reboot which nobody noticed due to the Cappuccino IDE being cached by the browsers
Where to Next
- now that the traffic has died down, we have upgraded our infrastructure and are getting this update out to our users and the HN community
- we're trying to get the word out that Akshell is a serious development platform and that we have developers lined up in Russia and elsewhere to quickly build webapps and API mashups for anyone who is interested
- we are keen to talk to companies with APIs that would like to use Akshell for promotion and running "demos"
- due to visa hassles, we are already booking the flights to the US, despite not knowing whether we will be accepted into YC, so if anyone wants to talk to us or meet when we are there, you can reach us at info@akshell.com
It looks like accidental launches are becoming a bit of a tradition. Thank you to everyone who helped us get here. Every Hacker News upvote and retweet counts!