148 karma · joined November 18, 2008
What it doesn't mention is that everyone who picks NoSQL generally acknowledges these downsides. So it doesn't do a good job explaining why people still pick NoSQL.
TCO is probably the biggest factor I would have in picking NoSQL. As a (Microsoft) SQL Server customer, we (developers) all hate the cost of scaling SQL Server for each customer. Each customer needs their own resource governance. This means either separate servers or a virtual machine, both tied to a SAN. Virtualization MSSQL licensing costs are large without much material benefit due to odd licensing strategies that negate most if not all financial benefit to virtualization for small IT shops like us. By comparison, the better we understand why customers use our product and the more engineering decisions we can make (get away with?) then the cheaper technologies we will be able to use (NoSQL, etc.).
I am curious if your hypothetical question is a reality. :)
10. Get into a rut early: Do the same process the same way. Accumulate idioms. Standardize. The only difference(!) between Shakespeare and you was the size of his idiom list - not the size of his vocabulary.
Silverlight is more secure than WPF, too, by the way, and has to be. For example, in WPF there is a very insecure static method that allows you to steal a bitmap of the entire screen! This was one of the first things taken out of Silverlight.
If you want to know more about Silverlight security, ask Nick Kramer who maintains the Silverlight security best practices document for Microsoft.
However, he has two technical mistakes in his post from a security perspective:
1) We don't know the attack vector. For example: If Mt Gox has a SQL injection vulnerability, then a sophisticated attacker will not waste their time doing a rainbow attack on a random user. Instead, if the account balance is not encrypted and the key kept secret, then the attacker simply needs to do a SQL injection attack that returns the account with the largest balance:
select top 1 t1.account_id, t1.balance from Account t1 order by t1.balance desc
Instantly, the attacker knows the largest balance. This automatically reduces the attack space. This is a standard trick attackers use to bypass even needing to guess a password.
2) Compounding this issue, it seems Kevin is right, the attacked account had a naive password susceptible to a rainbow attack. According to rumors, this attack was a pooled account that mediated all the assets traded on the exchange. This implies that MtGox used a password susceptible to a rainbow attack to secure the master account. To answer Kevin's question, what user would amass $8M in bitcoins and use a bad password? The system administrators. 0xDEADBEEF.
Postscript: I have never traded or owned bitcoins, or even signed up for an exchange. I just find the security breach fascinating!
Any time you comment out a WHERE clause on a big table while performing an aggregate report, Postgres chokes.
Then you can write a blog rant about how painful it is to refer to "the second-order (polymorphic) lambda-calculus with bounded quantification" as F-sub or F<sub><=</sub>
You do not have permission to call Direct3D directly. You can't even do cool hacks like you can in WPF, stealing the Direct3D video feed and writing it to a movie file. Everything is abstracted away by (underpowered) APIs.
All Secunia advisories on .NET Framework / Silverlight are presently patched, and the total number is relatively small compared to other technologies like web browsers and Flash. I don't really know enough about WebGL to compare, though.
My statement -- there is not a whole lot you can do to tune your queries -- is about rewriting the queries themselves, such as join hints, etc. Changing RAID configurations, organizing the data physically on disk differently, etc. have no bearing on the quality of SQL an ORM generates. So what does? Well, 20 years ago many DBMSes required you to compile stored procedures in order for the cost-based optimizer to cache anything. Today, not only does SQL Server 2008 R2 support fine grained plan caching and allow you to adjust the size of that cache, but you can remove from the cache any plan you dislike. You can also force a poorly performing query to use a specific plan cache. This is one of the ideas behind the LINQ Re-Motion project for .NET.
I am really arguing about where to put work effort into, and really NOT disagreeing with you. We are talking past each other.
I personally believe, as I wrote in the author's blog comments, that an ORM based on an algebraic model would likely be better than the big and irregular ORM APIs we have today.
The story is slightly different for open source RDBMSes which have inferior cost-based optimizers and multi-version concurrency control implementations that fail on a wider range of corner cases like:
select count(*) from Billing_And_Accounts_Receivable.Transaction
which is slow in PostgreSQL for reasons I'd rather not take the time to teach about here.
Compile-time type checking is very illusory with heterogeneous execution, not only for this reason but also because it is a distributed application and the schema can change.
I would be willing to help you write it if you'd like. I understand network disruption, partial failure, and weak/episodic connectivity very well, including scenarios where nodes in a network are required to remain radio-silent for long periods of time. Command-and-control military-style systems are much more general than the Web.
Related, automatic code distribution is a sort of Holy Grail in programming language design. It is difficult for many reasons. Even data is a huge problem. For example, if the toolbar buttons on Google Docs suddenly die (like they did for our marketing rep a few weeks ago), then the application can become extremely hard to use. Web applications don't handle this well today, and a cache doesn't really address this well, either, because it could be that the new resource invalidates an older cache. In other words, it is a version control and configuration management problem.
Put simply, offline applications need to support partial failure. The author of this article seems to want to write web apps without having to have much concern for partial failure. That is my own summary and take-away. If you agree, then we are on the same page.
Not dealing with partial failure means succombing to Fallacies of Distributed Computing - period. I wouldn't even trust the browser to do atomic swaps of cache', anyway, and I wouldn't even want to think about what that code would look like, or what the spec for the author's ideal would look like.
Bottom line: The author is defining words like "brittle" from one perspective only. What he wants is probably better provided by a library on top of caching. Moreover, I shriek at the idea of an application having to know its "size". That exposes the representation of the resource.
How can I get in touch with Andrew Binstock? (The author of this wonderful article.)
The reason LANs generally haven't cared about propagation delay is that they are not using networks already operating near the ultimate bottleneck: the speed of light. Some college campuses I know of have less bandwidth than an iPhone. But for Internet-scale computing, propagation delay is evident.
Wish I could've upvoted this 100,000,000 times.
As with anything else, assume a normal distribution of teaching talent, and hope you end up being taught by somebody on the right side of the curve.