The disk array error fiasco at VPSLink, has revealed some dark side of shared servers — or when the server hardware setup is out of your control. How would you know that your hosting company is using RAID 0 with SATA drives that is probably having double the failure rate as the drives are stripped?

Tough luck.

First of all, I am currently a VPSLink customer simply because of the unbeatable pioneer discount (<$13/month for a 256Mb OpenVZ VPS). So far the performance is quite acceptable for its price, and I can rest easy knowing that they are backed by a big hosting company (Spry.com) and not some garage operations.

However, although I’ve acknowledged that VPSLink provides budget VPS hosting, I was quite surprised that RAID 0 was used to provide performance and sacrifice reliability. Further more, the performance was actually not that great (VPS was often starving on IO), showing how many VPS they have packed into one server to bring down the price.

Fortunately, my VPS was not affected by this incident. A few things to think about though.

RAID 0 Stripping on VPS? Bad Idea

IO contention is quite often the bottleneck of VPS. However, I do not think using RAID 0 is a feasible solution to this problem.

With RAID Level 0, you basically split up your file system into lots of tiny slices, and then put alternate slices onto each of the two drives in the RAID set up. The advantage is speed gain in both reading and writing from/to the disk drives, as operations on consecutive blocks can be done in parallel.

However it is also at the expense of reliability. If one drive fails, you pretty much will loose the entire volume, as it is impossible to extract out files when you only have alternate blocks. And believe, disk drives do fail very frequently especially when you have a large sample. Even Google has no exception.

In web hosting world, loosing a drive in RAID 0 usually means extensive down-time — taking the server down, fsck the file system, scavenge the /lost+found, restore from backups, etc — which often translates to loosing customers. In fact, I would not host my websites on a host with less than redundancy provided by RAID 1, where there is always a hot mirror. With RAID 1 mirroring, you get only half the overall capacity. You still get the same reading speed, but writing will be only as fast as with a single drive. But I will rather write my files back slowly, than risking my entire file system fried.

How Storage is Handled at Your Shared Server?

That brings us to the next question — how is the storage handled at your shared server? With a dedicated server provider, it is easy. You tell them how you like it to be done! With my other two VPS hosts, SliceHost and GPLHost, at least it was stated clearly on the website that RAID1 is implemented.

However, there are many VPS and shared hosts out there who never mention their storage arrangement on their website. Some just say “RAID” (in VPSLink’s case). Some just give you a size in number of gigabytes. That alone should make you worried.

Backup, Whose Responsibility?

VPSLink claims to have nightly backups (from their RAID 0 drive array to a separate SATA drive), however they seem to have issues restoring the backups and applies incremental changes.

Sometimes I do wonder, should we rely on hosts to backup our servers for us?

The answer is a big NO. While the host is responsible for performing regular backups, it will be considered stupid not to have your own off-site backups. How are you doing it, it is up to you (I have briefly talked about it previously). However, make sure you do it! An earthquake can hit US east coast and sub-merge the entire Seattle into the Pacific Ocean — and you cannot tell VPSLink/Spry that they need to somehow find some divers to rescue the hard disks containing your data, can you?

And let’s face it. Web hosting is a commodity, despite denial from those in the industry. Not happy with the current host? Can’t get the site back up in time? As long as you keep your own sets of backups, you are free to just pack up and leave. Find another host, upload your files, migrate the DNS, and you are good to go.

Should I Stick with VPSLink?

Well. They are cheap, and I am getting the service I expected. Moreover I prepaid to June so I am not in hurry to leave before then.

However my VPS there is really under utilised — I am only running a few low traffic sites and cron jobs on it. I might as well get rid of it unless there is any new project that I need to off-load onto it. Not likely with my current schedule…

Anyone interested in buying a cheap 256Mb VPS off me that’s on a RAID 0 disk array? :)

Joyent’s new Bingo! on-line disk service gives you 100 gigabytes of disk storage on Sun’s amazing X4500 platform with 10 gigabytes of bandwidth per month over WebDAV for the super-low price of $199 a year! You get one user account and the ability to serve files from a public folder (for images, podcasts, whatever). On top of this, you get a complimentary 5-user subscription to Connector, Joyent’s amazing group collaboration product.

Pretty good price on a very good hardware let me say. I guess they must have sourced quite a few Sun Fire X4500 to make the price this competitive. At $16.60/month it makes a great online storage solution. However, monthly transfer has been limited to 10GB/month, so it would take almost a year to fill up your account without incurring extra costs.

Pricing is a bit funny though. First of all, isn’t it a direct competition with Joyent’s another service, Strongspace, which provides on-line backup via SFTP and Rsync? Sure, it gives you Rsync (instead of WebDAV), and there is no extra bandwidth charge, but at $8/month with only 4GB of storage, you can tell where the customers will go.

Moreover, Bingo! came with 5 user subscription to Joyent Connector, which itself costs $15/month, and comes with only 5GB of storage. I can see Bingo! is making the “Start” plan redundant.

Update: Take what David Young has said,

No, we’re not staring. We’re building. Bingo is a key foundation.

And what Brian Duffy has later commented,

Your pricing model is pretty cool… getting 100GB/mo with 10GB/mo transfer costs about the same as if you used Amazon S3.

I think I know what Joyent is building — an Amazon S3 killer. Note that Amazon S3 is not just an on-line storage with a cool API — it is going to be the foundation of other Amazon service products like EC2. Now Joyent has Bingo! backing Connector for storage, but I am sure there will be many others under their secret plans.

Moreover, the pricing is set to match with S3′s. Each gigabyte of storage comes down to 16.6 cent per month, plus included 10Gb per month transfer, and on top of that, the extra transfer is on par with S3.

API? Too bad Bingo doesn’t have a set of API like Amazon S3. When you have a well defined and implemented-to-death protocol like WebDAV, you don’t need an API :)

Well, that gets me excited.