Posts tagged deployment

When will people learn to backup?

Had another friend lose a hard drive today without a proper backup.  Pain!

I now have at least 3 copies of everything with staggered backups to different hardware.  For the digital signage software I manage there are two copies on the servers locally and another copy in the cloud on Amazon’s S3 storage which is itself replicated multiple times.

The basic concept that people need to use is “SHARED NOTHING”.

RAID is NOT the answer to data security, it’s a convenient recovery mechanism for failed hard drives but if your data is on two drives connected to the same RAID controller card on the same computer in the same room you have plenty of opportunities to lose it all.

Backups should ROTATE.  Backing up to the same location risks a failure in the backup that could wipe both copies, or copy bad data over good before the mistake is discovered.  For really critical data I have a daily backup, a weekly backup, and a monthly backup.  I also have two backup schedules, morning that copies to one set of drives on controller A and an evening backup that copies to a different set of drives on a different controller.

Why Amazon should get into the dedicated server business

Amazon ECS is a great way to scale an application quickly; being able to spin up 20 front-end web servers or 30 back end compute engines quickly is terrific and the cost savings from being able to keep them running for as short a time as you need are undeniable. But as I noted in an earlier post there are a few dark clouds …

In particular even a High-CPU Instance performs relatively slowly compared to a dedicated core 2 duo server. In our case we found it to be 30% slower. Many others have reported performance issues resulting from the slow cpu to disk bandwidth available on a virtualized disk system like EBS.

For tasks where latency matters or you need a better disk performance a dedicated server is simply a better choice. The snag is that that server has to be off in some other data center and you have to pass all your data back and forth over the internet to Amazon ECS or S3 you hit more slowdowns and you incur more bandwidth costs.

The solution is simple and I think it makes good business sense. Amazon should get into the dedicated server business. If Amazon offered dedicated servers hosted in its datacenters with free back-end bandwidth to Amazon S3 and ECS instances I predict they would pick up a significant chunk of the dedicated server market overnight.
With a hybrid approach like this you could put your database cluster on a few dedicated servers and spin up front-end web-servers and back-end compute engines as necessary without any compromises on latency or disk performance.