Comments on: Backblaze Vaults: Zettabyte-Scale Cloud Storage Architecture

By: Aleh Veraskouski

Aleh Veraskouski — Wed, 04 Mar 2020 12:13:35 +0000

Doesn’t “drives in the same drive position … are grouped together” open a possibility of loosing too many shards in case any of the vertical location related physical damage (like partial flooding of a data center or some fire)?

By: Sconi Gardener

Sconi Gardener — Fri, 30 Aug 2019 22:17:49 +0000

Does the process to recover data from a failed drive wait until the failed drive has been replaced by a technician or are there spare drives in the data center that can automatically begin the rebuild process?

By: Vishesh Gupta

Vishesh Gupta — Tue, 13 Aug 2019 14:22:46 +0000

Hats off to you guys! You guys have been very open about your datacenters, and its architecture. I am a technology enthusiast, and want to know more about the networking infrastructure of your datacenter. In the above article, you guys mention that each storage pod can receive 1Gbps data, and the vault in total can get 20Gbps.

1) Do you guys plan to upgrade to 10Gbps cards in each of your storage pods?
2) How does adding more vaults increase your bandwidth? Aren’t you guys limited by the speed of the ISP? Do you have a dedicated line for each of your vaults separately?
3) What is the network backbone infrastructure? How do you guys scale? I am really very much interested in the networking equipment used, and how you guys wire everything.
4) When a customer creates an account, and requests to upload data to your datacenter, where does that request first goes? Is the data broken down into 20 shards on customer pc, and then uploaded to different destinations, or is it done on a separate server (other than the vault) which breaks down into pieces, and then distributes it to the vault? In essence what is the first point of contact to the datacenter?
5) I have been really fascinated by CEPH FS, if i were to create something similar in a home lab environment with virtual machines, where do you think I should start with clustering?
6) Along with a storage cluster, do you guys also need a compute cluster / processing power to handle all the incoming data?
7) Since you guys have already revolutionized / economized the cloud storage industry, do you guys also plan to offer online rendering services / renderfarm?

Sorry for the noob questions. Just curosity.

By: Zyo

Zyo — Tue, 06 Aug 2019 10:50:33 +0000

In reply to department_g33k. fair point!

By: Jonathan Grove

Jonathan Grove — Tue, 09 Jul 2019 11:31:13 +0000

Look at those gorgeous storinators!

By: Brandon Bertelsen

Brandon Bertelsen — Thu, 27 Jun 2019 20:30:49 +0000

These posts are extremely interesting. Question, can you give a few examples of “rare expected failures” (vs. unexpected?)

By: oregondean

oregondean — Thu, 20 Jun 2019 12:39:17 +0000

11 9’s … sounds like a military drill team … congratulations … and on behalf of the many data owners out there – thanks!

You detailed the durability, relaibility and accessibility of how an individual file is stored and retrieved. I’d love to learn more about how the index/diretory to these files is handeld. It seems we have a fairly large collection of files stored on our local machines that facilitate this. How does Backblaze manage these indexes and how do you “find” the files we are looking for? Are these directory files backed up and managed the same as our regular files?

By: Jonathan Cormier

Jonathan Cormier — Tue, 18 Jun 2019 13:42:24 +0000

Whoops

90 zettabytes! (Also knowWithout changing

By: cryptochrome

cryptochrome — Tue, 18 Jun 2019 13:17:19 +0000

I am curious to know if your vault software is completely custom built or whether you are using (or building upon) one of the available distributed filesystems like Ceph or GlusterFS.