Finally, someone took time and said it. Moreover, I'd recommend for someone who didn't have exposure to "big iron databases" to look up performance benchmarks for Oracle, Microsoft and DB2, they're very popular not only among these companies but also among big-iron manufacturers like Sun, IBM and HP. It's crazy what they can do if you use them right, good luck building a business that will need such transactional capabilities of your DB.
I assume the real appeal behind SimpleDB is that you don't have to buy servers, disk arrays, backup systems, a data center, 24-hour techs to swap RAID drives, etc. You just put your data in there and forget about that stuff.
Exactly. And its not just hardware you can forget about. No longer having to worry about indexing, scheduled backups and database shrinks, temp table sizes, connection pooling, and all the other nasty stuff that goes along with maintaining a DBMS is equally as nice.
I've switched my project from a local MySQL instance to SimpleDB and I'm never looking back. It took some work to simplify my existing schema to be purely atomic (no joins), but after that you can forget just about everything else.
I would be interested in hearing more about your migration from MySQL to SimpleDB.
Basically, I am a big fan of normalization (since none of the other ways seem elegant to me). For example, if I were going to write the schema for a Blog, I would have an articles table, a tags table, and an article<->tag mapping table. This means that "show me articles with tag foo" is indexed (and "faster" than an O(n) search over the articles table).
The only approach I can see with SimpleDB is to tolerate scanning all articles to search tags (and authors, etc.) or to maintain a separate "document" for each tag (that links to the articles with the tag). Then I worry about concurrent updates to a tag ( {foo:[1,2]} gets updated to {foo:[1,2,3]} while someone else immediately changes it to {foo:[1,2,4]}; the correct result would be {foo:[1,2,3,4]} but adding 4 overwrites adding 3.)
Anyway, like I said, I would be interested in hearing people's experiences. Perhaps I am missing something obvious.
With SimpleDB, you can still accomplish many-to-one and even one-to-many. So in your example schema you described, you could have an attribute (aka column) in your tags domain (aka table) which holds the ids of the article items (aka rows) in your article domain that the tag is associated with. Since your data is denormalized though, you will have many duplicate items if a tag applies to many different articles. Thats ok though, SimpleDB was designed for this dont worry about large domains.
So this is how you could get your many-to-one mapping. To determine what articles have tag foo, you would just do the equivalent of select from tags where your attribute value matches the item id for the desired article.
It seems unintuitive, but you can actually do one-to-many mappings too with SimpleDB. So you could say "show me all the tags for article A" without a join table or duplicate items. The reason this is possible is because SimpleDB can attach up to 256 values to a single attribute. This obviously isnt possible in SQL. You just assign references to each tag id for that article to one attribute. Kinda like putting a list of Strings in a single cell in SQL. Be very careful with this though, as you cannot exceed 256 values bound to 1 attribute. So no more than 256 tags per article in your example.
But the problem with all this is keeping track of changes. updates and deletes do not cascade so you have to handle that yourself. And of course you have to consider eventual consistency if making frequent changes on top of each other.
My suggestion to help reduce the complexity of managing your domain references is to use this library (if you are using Java that is). It makes SimpleDB so much more usable: http://code.google.com/p/simplejpa/
I hacked together a SimpleDB-alike tonight and played with various index-on-write schemes. I feel a bit better about the situation now. Of course, now I just need to make it as scalable as SimpleDB :)
I don't want to bash something I personally not very familiar with, especially on this podium. But I don't know what the appeal of SimpleDB-like services is. I know why Google/Amazon needed them, but those reasons don't apply to 99.9% of startups.
You don't need to buy your own RAID/SAN and powerful servers to use RDBMS, nearly every modern scalable hosting provider will be happy to rent them to you just like Amazon: get an account on Slicehost for smaller projects and move up to EngineYard or Joyent for high-end stuff: they will all let you scale up as needed, use DB of your choice.
P.S. I also think the author forgot about yet another important advantage of RDBMS: there are tons of various tools for them. Yes, he mentioned import/export/backup, but the list is much longer than that.
True, but the reason that SimpleDB is so inexpensive is because all of the items you mention have gotten so cheap. A managed server, where the provider does all the backup and RAID stuff, costs ~$100/month. That's nothing, given the tremendous benefits (like, joins that allow for normalization, your own backup, no vendor lock-in, etc).
I understand the desire for something that's maintained by someone else and makes life easy, but if you're a tech startup, this stuff ought to be second nature.
