Moved IPVM To A New Server, Now 35% Faster

We moved the IPVM application / site to a new server [Update: the site is now 35% faster]

Here's a few technical details for those of you interested in server selection / specs (see previous discussion on picking servers).

Previous - Rackspace

We have been using Rackspace, though that server is originally from Slicehost, a Rails hosting specialist, set up in 2008 (by me) when we started. This server has 2GB RAM, 80 GB hard drive and 2 'vCPUs'. We paid ~$100 per month.

Rackspace has been fine for us, not great, but no major issues. The main motivating factor is that the server has accumulated 5 years of 'junk' from various changes and experiments (especially by me in the early days).

New - Digital Ocean

We have moved to Digital Ocean, which is a hot up and coming provider (see 'the Meteoric Rise of Digital Ocean'). Their main marketing claim is using SSD storage instead of hard drives. The other big practical differentiator is that, for the same price that we are currently paying, they provide 8GB RAM (4x more) and 4 cores (instead of 2).

Why Not a Dedicated Server?

We discussed going to a dedicated server but rejected it (both previous and new setups are virtual). The main issue is that if a catastrophic hardware failure occurred, we would be responsible for setting up our stack on a new machine - an issue that would be stressful and problematic both for us and for our members who would lose access during that time (easily hours, if not most of a day, depending when it happened). The main practical upside would be greater RAM / CPU but with Digital Ocean's 8GB / 4 core / 80 GB SSD setup, that's far more than enough for what we need even when we double traffic again.

Anyway, let me know if you have questions or thoughts.

The new server is live. It feels to me significantly faster and some single page tests confirm that. However, I'll know more when after a few days of traffic.

Also, one minor note - the server is now physically in NY instead of TX which might make a slight difference / benefit to East Coast visitors. That said, I am still in Honolulu and it still feels faster to me.

so this relates to "cloud video" and other things "in the cloud" where suddly the ISP's routing vaguaries can impact your business. Traceroute to the site looks fine, although the packets wander around inside Level 3's network in NYC a bit long as they finally figure out they need to go to Newark ("tracert".) "ping -n 100" looks solid from cloud hosts (in Fremont, CA and Chicago, IL). Kinda wild variations in time from the coffee shop I'm sitting in in Berkeley CA but I doubt that's you. DNS servers are sanely diverse. And I never noticed until I saw the post, it's not like I had an issue before you incentivized me to go all wine-snob about your hosting provider choice.

Ping tests from HNL to the ipvm server in NY/NJ range from 135 - 150ms. Traceroute shows a few hops in DC and NYC (but total time is like 10 - 20ms for that).

The big issues in web applications are typically client side/page rendering (javascript, images, etc.) or server side (database lookups, the application itself, etc.), not network. We've cut at least a few hundred milliseconds off the server side portion which is a huge gain overall.

The site is now 35% faster overall, which is a huge jump. Previously, average load time was a pedestrian 4.6s, now it is a fairly strong 3.0s. I confirmed this even after the spike in traffic from our newsletter today. I am happy about this.

That's a reduction of 5~ days loading each month for all members (250,000+ page views x 1.6s per page).

Ahh, yes, but what is the "real world" practical expected savings in load time? Remember, one 10MP camera does not replace 20(D1) cameras, or whatever it is.