Ask HN: Explain AWS • Hacker News Board

Re: Ask HN: Explain AWS

#31

Post by keefe » Wed, May 06, 2009, 4:56 PM UTC

jawngee wrote: post #27

keefe wrote: post #14

Earlier quoted context omitted.

After a cursory googling, it looks like AWS is still the only one where you can programatically modify the number of nodes in your cluster. If your goal is to launch a public facing site and automatically scale up to some limit as you get users, then AWS is a good choice. With AWS you are buying automated scalability and amazon's reliability (not that rackspace doesn't know what they're doing)

You can do it with Slicehost's API. http://articles.slicehost.com/assets/2008/10/3/Slicehost_API...

Interesting, I see those entries... I had been looking at this :

http://www.slicehost.com/articles/2008/3/26/announcing-the-s...

". Please note that the current iteration allows access to DNS only."

I guess they had another iteration since then, thanks for the info!

Re: Ask HN: Explain AWS

#32

Post by delano » Wed, May 06, 2009, 5:02 PM UTC

psranga wrote: post #23

mindhacker wrote: post #21

Earlier quoted context omitted.

I've used EC2 for running web servers as well as database server with the db servers being replicated to rackspace for backup. I have always wondered about how EBS as disk will affect database as it is essentially a network drive. Any info on that would be great.

Yes, does anybody have data on latency and throughput of EBS "disks"?

I ran a few preliminary EC2 disk benchmarks in September: http://solutious.com/blog/2008/09/23/ec2-disk-performance-li...

The Solaris performance was really poor compared to Linux. I emailed Sun and initially they were really receptive but never explained what was happening. I'm hoping to re-run the tests again soon to see if the latest OpenSolaris (2008-11) has improved.

Re: Ask HN: Explain AWS

#33

Post by keefe » Wed, May 06, 2009, 6:42 PM UTC

latortuga wrote: post #13

keefe wrote: post #5

Each EC2 instance is a virtual, private server which looks just like any other linux box. However, files on disk are part of shared drives and are not guaranteed to persist. You have three choices : attach and pay for an elastic block store that IS persistent, make periodic backups to S3 or forgo the use of the disk altogether and use SimpleDB instead of mySQL. I'd recommend using an elastic block store.

I'll back this up with the addendum that SimpleDB has useful applications where using it might be a better choice than MySQL (e.g. a service that gets lots of reads but real-time updates isn't necessarily mission critical)

I think it's also interesting to note that once you do a cluster, you have a distributed system and instantaneous writes are no longer a cheap operation... Plus how many operations do you need to access full data in your database? For simple one off access, you could propagate the changes into ram on each of your cluster machines (memcached) and only relatively "unimportant" search operations would see stale data.

Re: Ask HN: Explain AWS

#34

Post by mattsmall » Thu, May 07, 2009, 4:11 PM UTC

RightScale's paid edition has prebuilt MySQL Master-Slave configurations with or without EBS. EBS is great if your transaction volume is average, but on high IO disks it can get prohibitively expensive. The RS Manager for MySQL has failover, recovery tools as well. All you need to do is plug in your DB dump.

RightScale's Developer Edition is always free - AND - RightScale Website Edition is FREE for Ycombinator startups while you're actively engaged in the program.