As I mentioned recently, I moved my hosting to Digital Ocean (link is referral code). All these comments apply to a 20GB instance I have operated in the nyc2 region for about 2 weeks, the sum total of my experience with Digital Ocean.
My needs are modest: I run this personal blog (python), a moin wiki, and some miscellaneous other web-facing stuff. I've selected the most basic 20GB disk / 512MB RAM / 1 core droplet size, which costs $5 a month.
Overall I'm happy with what I'm getting for the price I'm paying, and my sites are substantially more responsive than when I operated them on dreamhost. But there are some caveats:
- The available graphs (instantaneous transfer, CPU, and disk) are unreliable -- sometimes it is hours or even days between updates, sometimes multiple updates per hour (based on communication with DigitalOcean, this problem may be worst in the nyc2 region)
- "Real Time" graphs never seem to be realtime. If you want to look at a fresh graph, you have to reload the page and then click "graphs".
- Graphs aren't available via API. Discussion of graphs here
- I/O, Bandwidth and pystones are somewhat variable
- Bonnie++ zcav gives 579MB/s (st dev is 40% of mean) over a 20GB /dev/vda—apparently not data-dependent compression, fast spots move around from run to run
- Other performance measures vary as shown below (now based on 818 hourly tests):
statistic mean std dev Units pystones 103229.1 ± 15757.1 (15.3%) Pystone/s ping 14.9 ± 1.1 (7.3%) ms down 24.5 ± 9.0 (36.6%) Mbit/s up 49.7 ± 29.6 (59.6%) Mbit/s
- Apache file transfer as measured by 'ab' from remote is not as fast as speed measured by speedtest-cli, but that's probably my own lack of tuning
- Contrary to my expectations, there was not a clear daily or weekly component to performance variations.
Update, December 25, 2013: An earlier version of this page was inaccurate, particularly about pricing. An e-mail from digitalocean clarified that the price of the small droplet is capped at $5/month (and so on for other droplet sizes), so (now-deleted) my calculations based on per-minute pricing were wrong. They also clarified that IP addresses are never changed for a droplet as long as it is active, and that the degree of performance variation I measured is more or less what was expected. I also deleted some items about bandwidth quotas and monitoring because apparently no bandwidth charges are being made at present.
Update, January 9, 2014: Update statistics based on 818 hourly tests. I got tired of getting the hourly results in my mailbox, so that's the last update I expect to do.
I'm trying out a different, VPS hosting provider, DigitalOcean (link is referral code). I expect the performance to beat DreamHost shared hosting and the price to maybe be lower (if the 20GB instance is enough storage for me). On the downside, DigitalOcean doesn't currently offer IPv6. If you see this post, you're talking to the new machine.
All older entries
Website Copyright © 2004-2017 Jeff Epler