Hacker News Reader

Comment by btmorex | Hacker News Reader

Show me one cloud offering that gives consistently good or even average disk performance. (hint: there isn't one)

Basically, if you're running a database that does not completely fit in memory you should be on dedicated hardware.

I'd also point out that a lot of advantages that people routinely cite as cloud strengths are more about cloud vs traditional hosting or colocation as opposed to cloud vs a place like softlayer. softlayer can provision a custom build in a few hours (yeah vs. minutes, but who really cares that much) and you pay month-to-month without a contract.

View on HN

Show me one cloud offering that gives consistently good or even average disk performance.

You mean like newservers.com, SoftLayer Bare Metal Cloud, stormondemand, or one of the other metal clouds?

I believe the orionvm cloud would qualify as "good or even average disk performance" (http://orionvm.com.au/blog/3rd-Party-Performance-Benchmarks/ Benched by cloudharmony.com), as they are 60% faster than a dedicated server with 4 * 15k SAS disks.

Disclaimer: I'm a director at orionvm.

I run a database that does not fit in memory on AWS with great success. Generalizing is dangerous. If you work with and understand the constraints imposed by the environment you can do some pretty amazing stuff.

I have found Linode disks to have the best performance in all

http://pastebin.ca/2049137

You are correct, I/O is the challenge in administering systems in a virtual environment. My database which does not fit in memory does fine on a high-load site because I cache it responsibly. For comparison, here are awful results from a new player called ChunkHost, who I signed up for with the purpose of testing

http://pastebin.ca/2049142

The sequential write throughput there is troubling. This comparison from a couple years back is interesting too

http://journal.uggedal.com/vps-performance-comparison/

I've linked this URL before but it really does the best of breaking it down. What cloud providers have you tried? In my experience there are vast gaps between certain ones, Amazon no exception. Hard to stereotype cloud with gaps like those.

Even if SoftLayer could provision me a new box in ten minutes the improvement to my sleep from not waking up for every disk failure and submitting a remote hands ticket at who knows how much per pop far outweighs anything else.

I have also had good experience with linode performance (just a small instance I use for a few personal projects). However, AFAIK linode is just using local on-box disk, which is a whole different animal from EBS.

To say that a database has to fit entirely in memory to achieve good performance is a ridiculous proposition, and simply shows you have zero actual knowledge of modern database server internals or administration. Countless sites happily serve oodles of pageviews per day with actual memory usage far below the disk space used by their databases. Hint: they're not swapping, either.

In general, if you really believe what you're saying, you either (1) have a very poorly designed application, (2) have a very poorly designed database environment, or (3) are speaking to a specialized application that wouldn't reflect the majority of environments operating in real life. This isn't to say it isn't a combination of these options, mind you. I didn't even start on utilizing caching in applications, because it's clear there are other hurdles to overcome first.

I don't think he is saying that. He is saying that in a non-dedicated environment, you share the same spindles with other tenants who may have different I/O access patterns than your application. Careful choice of indexes, good data locality for fast reads, making sure writes are sequential - all that goes out the window if some other application is causing the disk to seek all over the place.

Reply on news.ycombinator.com