EC2, load balancing and fault tolerance
So we’ve decided to switch to Amazon’s EC2 for our hosting. Not that we’re unhappy with the service that Bytemark have been providing for our cluster, but EC2 offers a degree of flexibility, and hardware fault tolerance that you just can’t get when you manage the underlying virtual instances yourself. We’ll be making the switch for most of our hosted services just after Christmas.
The one thing that EC2 could do with is a fully fault-tolerant solution for clustered websites. We’re not able to truly factor out the single point of failure when load balancing. We’re getting round this by keeping another load balancer up and running as a spare, and automatically remapping the IP addresses should the first load balancer go down.
This will cause a maximum of a few minutes outage in the event of any server hardware failure, which is better than our existing setup, but it sure would be nice if EC2 handled this for you…
Having said that we’re pretty happy with the way EC2 works. I’m a full believer in the concept of Utility Computing – eventually computing power could well be provided in much the same way electricity is today…
In Stratus Technologies’ world of mission- and business-critical application support, fault tolerant and clusters are rarely seen in the same sentence, the exception being to say that clusters can never be fault tolerant. Highly available, but not fault tolerant. FT and clusters are distinctly different architecures … the former designed to prevent outages from occuring, and the latter for fast recovery after the fact; the consequences for clusters are unplanned downtime, application restart, and possibly data loss.
On the subject of load balancing, why not get the highest availability while not getting caught in high prices? Kemp’s got some great load balancers that are low priced and high in quality:
http://www.kemptechnologies.com/?utm_source=blo...
Hi, what was up with Bytemark's service? What snags did you hit? Thanks
I was never been aware of EC2 before but it seems to have a lot of usage and purpose.. thanks a lot for the info. I'll check it out.
Hi,
We switched near the beginning of last year from EC2 to bytemark, due to excessive costs from large and extra large compute units at EC2 (Currency and unpredictable costs did not help). We have had a number of outages from Bytemark, however I would be curious to hear more on your experiences and am wondering whether you are switching because of the buzz around cloud computing?
I do find getting answers from bytemark to be challenging and considering we are spending a fairly large amount on servers, I find this unacceptable. In addition, the documentation on their site is lacking.
I’m writing up some fairly detailed documentation on high availability, and will publish this some time this week, which you may find interesting.