A blog with detailed instructions on building a custom storage pod for a data centre has generated enormous interest online since it was posted on September 1.
The article, by Tim Nufire at online backup service Backblaze, contained exploded diagrams and wiring schematics that the company used to build a rack-mounted unit containing 67 terabytes for just US$7867.
Backblaze CEO Gleb Budman said the response to the article in the past 36 hours had been "a little crazy".
"We literally had no idea what to expect. We went down this path because we needed cheap reliable storage, and we never intended to build our own storage - that's not our core business. And then we thought maybe we should share it with others."
Budman did a Google search on the pod and found that in just a day and a half it had generated 30,000 hits.
Backblaze spends "hundreds of thousands of dollars" on storage to support its online backup business, which targets individuals and small business.
Budman said the company had decided to build its own storage units because the cost of assembly was up to eight times lower than the cheapest product sold by IT vendors.
"When we looked at [expanding storage] for ourselves, I could go the open market and buy a 1.5TB drive for $150. Storage vendors would charge $1000 for a 1.5TB drive to put in a storage system," said Budman. "A lot of it is just pure mark-up."
In his blog post, Nufire compares the cost of assembling a petabyte of storage using Backblaze's storage pods to the retail prices charged by vendors for the equivalent amount of storage.
The closest vendor to the Backblaze system's cost price of US$117,000 is a Dell MD1000 with a sale price of US$826,000. A Sun X4550 retails for US$1 million, and the EMC NS960 is priced at US$2.86 million.
Budman added that this was not a direct comparison, as the figure for the Backblaze storage pod was only the cost of building the box.
"Obviously there's no tax, there's no shipping, there's no mark-up for any sales team or marketing team, or the support infrastructure or warranties. We have just published the cost to build it," said Budman.
Budman also pointed out that the Backblaze pods were designed for storage only and not for high-performance transactions. "We don't run databases on them," he said.
The storage pods had a throughput of an "easily sustainable" 200Mbps. "They could burst higher than that. But we don't need any one of these boxes to accept crazy amounts of data at any one time," said Budman.
Budman said that the higher prices charged by IT vendors were "a combination of value and mark-up".
"But for a lot of companies out there that just want to store a lot of data I think there's a lot more mark-up than value," said Budman.
Budman said that a company could probably cover the cost of support and marketing by doubling or tripling the cost of assembly. Even tripling the build price to US$351,000 would still make the pod less than half the price of the next cheapest solution.
"It's still not 8 or 10 or 20x the cost," said Budman.
Read on to the next page to find out why the company replaces one drive per week.
The online backup service provider formed in January 2007 and launched in the middle of last year using with a Dell system. Backblaze then experimented with an array of external mini-NAS boxes connected to a central computer to handle its expanding storage requirements.
"It's been quite a rocket ride. We launched the Windows version in September 08 and we launched the Mac version in April this year," said Budman. "We added a petabyte for our customers in the last six months. Facebook's entire photo storage is a PB and a half, and we've had that for the past year."
The company gives individuals and businesses unlimited backup for US$5 per machine per month. A competitor, EMC-owned Mozy, also gives users unlimited backup for the same price.
However, this offer is restricted to home users and is also throttled to prevent uploading large amounts of data, said Budman.
The company's business model requires cheap storage to make a profit on its unlimited backup offer. Budman realised he needed to build the storage units to make the business viable.
"We couldn't make a profit on it if we had bought the equipment. We wanted to offer a fixed price because nobody really knows how much data they need to backup. We didn't want people to worry about whether they had enough space," said Budman.
"We are one of the few companies for whom this was a necessity to do. There aren't a lot of companies that have the scale requirements in needing petabytes of storage and also have the cost pressures, where it's critical to have it cost effective."
The company committed to building its own pod and drew on support from outside the company to design the system.
Budman said the process was not as easy as they first thought. "We thought that hardware was like building blocks, like lego, and you put it together and it just works. But it took a while to get right."
Backblaze has been using the pods for a year which provide a petabyte and a half of storage. Reliability issues have been ironed out; Budman said the company replaces one drive a week in the whole set-up.
None of the pods themselves have failed, he said. "It's only been a year but the pods themselves are brand name components for the most part and they've been solid."
Backblaze discovered that the type of drive made a difference to the pod's operation. It has standardised the pods on 1.5TB Seagate drives because they offered the best combination of price, performance and reliability.
"We expected that one drive would be the same as another, but we found that some drives performed better and some worse," said Budman.
Read on for why Backblaze chose Debian as the operating system.
The hardest part of designing the pods was trying to get all the pieces to work together without having the drives pop out of the Raid 6 arrays, said Budman.
The storage pods use Debian Linux for their operating system. Backblaze tried several versions of Linux to find compatibility with components but otherwise the OS needed little configuration. "Once we got it right it was fine," said Budman.
Backblaze has a "tonne of our own software" running on the pods to encrypt, deduplicate and assign data, which was "incredibly complicated" to write, said Budman.
Backblaze was releasing details of the software stack but withholding its own proprietary software from the public.
"We wanted to release enough so that someone could make the pods and write data to it over the network. Then it becomes a general purpose device that a VAR could install in their data centre and could be used in various shapes and sizes," said Budman.
The company has had many requests to buy or resell the pods and Budman is working out what Backblaze does next. He has considered making one-off sales but said that it was not the company's focus.
"We don't like to say no," said Budman. "But to be totally frank we are busy. Our core online backup business is growing quickly and we are pretty heads- down focused on growing that. We didn't really plan for a path where everybody wants one."
Backblaze has an affiliate program with 150 partners. The affiliate receives an ongoing share of revenue for each user it signs up.
A reseller version is in the works which adds custom billing and a management portal that lets the reseller view the status of customers' backups. Budman said 10 to 20 resellers are contacting him every week.
"We've had 70 sellers in the past 24 hours sign up to be resellers of the core online backup service or these storage pods," said Budman.
Some readers expressed an interest in making the pods for their own companies. "We had a lot of people saying we are using X vendor's products internally but we are going to try and experiment with this."
For more information see https://www.backblaze.com/partners.htm.