Three bits of Amazon S3 news:
- We’re now using S3 as the primary storage for WordPress.com, rather than just for backups. We have some layers in front of it, notably Varnish, so the majority of our serving doesn’t hit S3. Still, our AWS bill went from around $200/mo to $1500/mo, and rising. It has simplified some of our requirements, but doesn’t look like it’ll save any money.
- Amazon now has a Service Level Agreement (SLA). Big companies like this, but in the real world I’ve found there to be a low correlation with service reliability and the presence of a SLA.
- In the Amazon newsletter they promoted Content Spooling Network as a good use of their services. Unfortunately, the service appears to be tailored for using Mechanical Turk to “ghostwrite keyword-based articles for SEO,” or more succinctly, “spam.” Get a web-savvy editor for that newsletter, guys!
A $1300+ increase in monthly costs? Is it really worth that?
Akismet has caught 232,946 spools for you since you first installed it.
Heliologue, for storage costs they’re hard to beat, but their bandwidth is vastly overpriced.
wow … big jump in that price … but I think in the long run it will help you guys!
Darin
Matt, I really respect the transparency you are providing around your financials. Awesome.
I am honestly surprised by the assessment that you are not saving money, especially given SmugMug’s now famous report on how it has saved them money. At the rate WordPress.com must produce content aren’t you constantly adding more storage? And doesn’t that hardware degrade and depreciate over time?
You are really losing money right? Your AWS bill may have gone up, but surely you are now saving money in other areas right?
Terribly Curious,
Byrne
I wonder how you guys managed to have S3 working as primary storage.. some well crafted plugins, some hyperdb-like magic, or something else?
We was thinking about S3 as our new primary storage server (we are serving actually more than 5000 slightly more than 4 or 5 GB) but we stuck on the implementation, and decided to work on this later this year… I’d love to know better some insight, just to not reinvent the wheel 😀
Apart from that, I think that wordpress.* is often a notable resource for testing and promoting technologies, and let even us little coders know something new.
This community workflow is just amazing. Keep up this good work, and thank you!
Well on the server costs side, we’ve gone from 1 server per datacenter costing 2010/mo total to 4 cache servers per datacenter costing 2236/mo total, plus the S3 cost which looks like it’s going to be about 2700 this month, so net it’s going to be about 3k a month more expensive.
However we no longer worry about storage capacity on the file servers, we have way more redundancy and I/O in each datacenter (for hot files) with no complicated cross-datacenter failovers, and maintaining the cache servers is easy. They’re quickly interchangeable. I feel that overall user file serving, which is a pretty crucial part of our service, is overall more reliable. I’d feel more comfortable doing things that increase storage usage now.
So right now it’s costing more than double what things used to cost, and is slower for uncached files, but I think as we tweak the caching we’ll get a higher hit rate and get the bandwidth cost down, which is 95% of our S3 cost.
AWS is great for some things, but bandwidth is not one of them. For high-bandwidth purposes you’re generally much better off with something like serverbeach.com, or whatever other host you like who offers you 2 bazillion TB per month transfer for only $199 or whatever.
If anyone’s interested in using Amazon S3 with their own hosted version of WordPress, I have a plugin available to help do just that:
http://tantannoodles.com/toolkit/wordpress-s3/
I’m surprised you don’t use Joyent’s new accelerators. I thought you WordPress folks had some kind of relationship with TextDrive/Joyent. They recently had a post about how their service is much cheaper than S3 for heavy-volume accounts like yours. You might want to do the math and compare, I bet you’d save money.
Good times…
I’m with Paolo on being curious as to how you achieved this.
I’d like to know more about the technical implementation, what plugins were used, code changes required and whether there are exceptions to the rule (are all WordPress static files also being stored on S3, etc).
Namely, if someone wanted to achieve the same where should they start and what pointers would you give?
I signed up with Amazon aws in October and uploaded my podcasts around the 20th of October. I was shocked to receive a bill of nearly $900 just for two weeks usage on 1st of november, I’m doubtful that my lessons were downloaded or viewed nearly quarter of a million times.
In the end, I learnt it the hard way, Amazon s3 is rubbish and i wouldn’t recommend it to anyway. Reporting & statistics is crap, customer service is not good (no telephone & i emailed them twice and still didn’t get a reply) Plus i don’t think they protect against spam which i believe is the cause for the majority of bandwidth.
Guess i learnt the hard way, not to use pay as you go like S3. Unlimited bandwidth is the future. I had to delete all my files from s3 and shut the site temporarily. nearly caused me to go bankrupt
I think im going to rethink my plans on using s3 now…