Notes on Backups

My normal routine for data storage and backups has been ad-hoc and stupid. I use Windows File History and CrashPlan for Mac to backup computers to a home desktop. Other important data is stored on various cloud services. Unfortunately my 8 year old machine is starting to fail, so it’s time to solve this issue.

I considered using an online backup service like CrashPlan or BackBlaze. It’s $60/year for 1 computer, or $150/year for 10 computers at CrashPlan. The problem is home internet connections are slow and potentially limited. My cable modem is 150Mbps down, 10Mbps up, but I usually get less than that. This means I can download 1.5TB per day, and upload 100GB per day at best. To initialize a 1TB backup would take 10 days. Furthermore, Comcast has threatened to limit monthly usage to 300GB. This would make online backups impossible.

The alternative is a home NAS. A 2-disk NAS running RAID 1 should suffice. The price for a diskless NAS ranges from $100-$200 from WD, Synology or QNAP. They can do all the backup and file synchronization tasks a typical home user would want.

Since my desktop is dying I’d prefer a “SOHO” NAS with more features. Some NAS systems can run virtual machines; i.e., it’s a small computer. The QNAP TS-251 can run VMs and containers, act as a print server, and has lots of apps. It has HDMI, so it can be used as a computer. For $300 diskless, it’s $100 more than a home NAS. I’ll likely get this one.

Almost any hard disk will do. The reports on hard drive reliability from Google, BackBlaze and CMU basically show that hard drives are commodities. Google says sometimes you just get a bad batch, like BackBlaze’s problems with Seagate’s 3TB drives. For a home user the fancy enterprise features don’t matter. HGST gets very good reviews, but they cost more.

The final issue is backing up offsite for extra reliability. Of course, the issues with online backups remain. While S3 sounds like a solution, for 1TB it costs $12/month (infrequent access) to store and $90 to download. Even Backblaze cloud storage is $5/month, $50 to download 1TB. If my whole NAS is destroyed in 1 year, to restore from S3 is $234, from Backblaze it’s $110. It’s absurdly expensive. One caveat, perhaps, is you might not need RAID if the cloud provider is already doing it. Instead, you could use a small RAID partition temporarily until the data is uploaded. Then move the data to regular disk. For a 1TB RAID 1 on a pair of 4TB disks, you can use 7TB of disk (1TB mirrored).

A cheaper solution is to mirror a drive at someone else’s place. You only need to buy an external drive and stick it on their network. If the NAS fails, your friend can FedEx the drive for a modest amount. Bandwidth is still limited to upload speed (100GB/day). This means the backup is not “consistent”. When you write data to the NAS, there is a potentially long delay before it’s replicated offsite. When the NAS fails any data pending for replication will be lost. This is true for any online backup strategy. Why aren’t more people concerned about this?