unlimited cloud backup for large data sets for linux?

[BLinux]

I'm looking for a 3rd backup location and have started to consider cloud backup providers. I have a somewhat large data set, about 6-7TB. It resides on a CentOS7 server, so whatever solution I pick needs to support Linux.

A friend told me he uses CrashPlan, and they have a Java client/server system that works on Linux. I'm trying it out right now, but it is super slow and I've tried a variety of suggestions i found online (set de-dup to minimal, set WAN de-dup max size to 1 byte, turn off compression, reduce frequency check to 1 day, change central backup servers - have done this 8 times now) to speed up the upload. I'm getting less than 1Mbps most of the time. A 1TB is going to take like 80 days, so it's going to take 1.5 yrs to upload my data set. I know there's a seeding service, but at this rate I don't think it can even keep up with daily or even weekly changes.

One of the things I want to backup are the VM qcow2 disk images, which are used by live servers running all the time and hence constantly changing. Each disk image is at least 10GB and many are larger. And there are about a dozen to two dozen VMs running all the time.

I currently have an offline backup system already, which is my primary system. (its a bunch of external hard drives that are connected to a managed outlet that gets powered on during the backup, which takes a snapshot of the VM disk images and powers off the disks) I'm looking for a 3rd backup solution for "offsite" backup.

Any suggestions? what are people using for cloud backup for large data sets?

 

[whitey]

I've heard good things about rsync.net...never tried them though, several here have though and have reported good results.

 

[Patrick]

What I would do in this case is build a machine as a backup target and Colo it. Or get a dedicated server with a lot of storage.

That is actually a planned item for my FreeNAS series.

 

[BLinux]

I've heard good things about rsync.net...never tried them though, several here have though and have reported good results.

I just looked it up and it's about $1000/month for my 7TB!! $12000/year is orders of magnitude more than I spend on my primary backup, so I don't think this makes much sense for me as offsite backup.

 

[BLinux]

What I would do in this case is build a machine as a backup target and Colo it. Or get a dedicated server with a lot of storage.

That is actually a planned item for my FreeNAS series.

that's an interesting thought, but colo with enough bandwidth would cost more than I think I want to spend here. however, you just gave me the idea of leaving a NAS server at a friends house and if we both have Cox's "gig life" service, transfers should be really fast. I would just need a way to remote power up the NAS, do the backup like once or twice a day, and remotely power off the NAS so as to not commit my friend to a higher electric bill. I guessing it'll cost about $2K-3K in equipment (hard drives, server, network managed outlet, a Pi-like low power device to VPN/remote connect so I can control the outlet, etc.).. still a lot better than $1000/mo and hopefully get transfers in the x00Mbps range.

anyone done something like this with a friend? just wondering if there are any unforeseen issues I need to consider...

it's still a lot more than the $5/mo CrashPlan that I was hoping would work out....

 

[whitey]

I just looked it up and it's about $1000/month for my 7TB!! $12000/year is orders of magnitude more than I spend on my primary backup, so I don't think this makes much sense for me as offsite backup.

They will do 6 cents per GB if you are a knowledgeable user of ZFS. Just got off the line w/ them. Discounts get even better as you scale out stg down to 3cents per GB for 10's of TB's. They use beehyve so multi-tenancy is assured.

 

[BLinux]

They will do 6 cents per GB if you are a knowledgeable user of ZFS. Just got off the line w/ them. Discounts get even better as you scale out stg down to 3cents per GB for 10's of TB's. They use beehyve so multi-tenancy is assured.

thanks! that's good to know, but $430/mo is still more than I want to spend for this... bare in mind, this is my home server.

 

[EffrafaxOfWug]

I built a NAS box for a mate and keep a set of my offsite backups on it. rsync over ssh over an IPsec tunnel. Most of the data is static but I've got three VHD disc images from weekly windows backups (plus a bunch of small xen disc images from little utility VMs) that transferring via rsync makes very bandwidth efficient (although I had to write a special sync script around the windows backup utility). Total footprint from my stuff is about 10TB, he's got ~1TB of stuff on there that I also keep a copy of on my side.

Bad side is you need a mate/family member/gullible member of the public who's willing to host a server for you and you need to pay the money up front, but it's much cheaper than the £$€pcm a hosted solution would cost me.

 

[whitey]

thanks! that's good to know, but $430/mo is still more than I want to spend for this... bare in mind, this is my home server.

I'm w/ ya, I would NEVER spend that kind of money for this purpose...as others have suggested build a 1U box and stuff it w/ 4tb seagate drives and a couple of ent class ssd's for accel and you're GOLDEN! 14 TB usable on a sc113 in raidz2 config...nuff said

Now you just gotta find a sucker to plop that box at their house. If you do a Xeon-D (2-4 core) build and toss them $200-300 up front for yearly elec cost should be an easy find though. Hell send it to me at that rate. Plenty o' room in the HP 1/2 rack here still :-D

I wouldn't mess w/ on/off...on/off shenanigans though, 24/7 or bust!

 

[Jon Massey]

B2 from backblaze has been very good for us, rclone is a good client. Currently got 16TB stored with them at a very reasonable price (0.5c/GB/mo). They'll post you a hard drive with a snapshot of your data if you pay for the drive.

 

[EffrafaxOfWug]

I'm w/ ya, I would NEVER spend that kind of money for this purpose...as others have suggested build a 1U box and stuff it w/ 4tb seagate drives and a couple of ent class ssd's for accel and you're GOLDEN! 14 TB usable on a sc113 in raidz2 config...nuff said

Now you just gotta find a sucker to plop that box at their house. If you do a Xeon-D (2-4 core) build and toss them $200-300 up front for yearly elec cost should be an easy find though. Hell send it to me at that rate. Plenty o' room in the HP 1/2 rack here still :-D

I wouldn't mess w/ on/off...on/off shenanigans though, 24/7 or bust!

If it's solely for backup, I would suggest that that's waaaay overkill... even if your mate's also using it as a file server (say, with access to one of the dirs where you keep your linux ISOs) they're unlikely to need full-fat SSD caching to get decent enough performance. Also I'd steer clear of 1U and for go for a desktop style case if you can fit the right amount of drives cases with four hot-swap bays are pretty cheap, plus a cheap duallie pentium (still fast enough to keep up with rsync, ssh and LUKS discs)... or go for a cheap HP microserver and you could get the same even cheaperer. Desktop formats IMHO tend to be easier to hide in a cupboard and much quieter than the 40mm screamers commonplace in 1U's...

MAF (Mate Acceptance Factor) with this setup was high enough and power consumption low enough that he didn't want money for power, especially because it also gets him free offsite storage as well. The odd few rounds down the pub from the money saved doesn't hurt either (Pint Acceptance Factor, or PAF), and serves as the perfect opportunity to swap portable HDDs if we need to do a big sync for any reason. Extra peace of mind in having offsite backups plus money saved also results in higher WPSOAF (Wife/Partner/Significant Other Acceptance Factor) also.

 

[talsit]

If it's long term backup, why not look at AWS Glacier? It's cheap long term storage.

You pay for bandwidth, getting the data on storage and then a low monthly rate to leave the data there.

There are clients that will update your changed data only on a regular basis, again, you pay for transit and disk access.

They have a calculator somewhere to give you an idea of monthly cost.

 

[BLinux]

B2 from backblaze has been very good for us, rclone is a good client. Currently got 16TB stored with them at a very reasonable price (0.5c/GB/mo). They'll post you a hard drive with a snapshot of your data if you pay for the drive.

thanks for pointing out rclone!! I had considered backblaze in the past, but their normal consumer side product team refuses to release a Linux client. I didn't look at B2 and didn't realize there was a client for Linux that can work with it. It looks like it comes out to $5/TB/mo roughly... which is reasonable, but still $35/mo for 7TB. That is reasonable, but i guess I was initially hoping Crashplan would work and was excited about the $5/mo for unlimited... so now i have to adjust my internal expectations.

 

[BLinux]

If it's long term backup, why not look at AWS Glacier? It's cheap long term storage.

You pay for bandwidth, getting the data on storage and then a low monthly rate to leave the data there.

There are clients that will update your changed data only on a regular basis, again, you pay for transit and disk access.

They have a calculator somewhere to give you an idea of monthly cost.

Thanks for pointing out Glacier. It's interesting, but with the mention of B2 @ 0.005/GB/mo, I noticed the cheapest I can find Glacier in US regions is 0.007/GB/mo so I think B2 wins on per GB/mo. And Glacier I have to pay for bandwidth, is that right? B2 seems to be free on bandwidth...

 

[talsit]

Yes, bandwidth in and out. It works from me but I only have 1.2TB there right now.

I keep live backups on Crashplan (~3TB total) and a local separate array and I do an incremental backup to Glacier once a month.

It took about three months to do the initial Crashplan backup. If I have a need for a full restore I can get it off Crashplan or (for a fee) they will ship me a drive with my full backup on it so I can do a local restore.

 

[BLinux]

Yes, bandwidth in and out. It works from me but I only have 1.2TB there right now.

I keep live backups on Crashplan (~3TB total) and a local separate array and I do an incremental backup to Glacier once a month.

It took about three months to do the initial Crashplan backup. If I have a need for a full restore I can get it off Crashplan or (for a fee) they will ship me a drive with my full backup on it so I can do a local restore.

how did you get 3TB up to Crashplan in 3 months???!!!! if I could do that, I would try to stick with Crashplan... the upload rates I was getting would be 1TB for every 90+ days... if I could get my 7TB up in Crashplan in 7 months, I would consider it... see my OP for what I've tried to speed up the upload. can you tell me if I missed something?

 

[talsit]

I just used the default settings and didn't really look at it again. It was several years ago and I had DSL so I didn't expect a lot of speed.

 

[BLinux]

I just used the default settings and didn't really look at it again. It was several years ago and I had DSL so I didn't expect a lot of speed.

and in spite of your DSL slowness, you were still able to get 3TB in 3 months? I have 10Mbps upload speeds from my cable provider, and right now i'm getting 800Kbps... it's stupid slow.

 

[talsit]

As best I remember... I started in the late summer and remember it finishing up around Christmas.

 

[Jon Massey]

thanks for pointing out rclone!! I had considered backblaze in the past, but their normal consumer side product team refuses to release a Linux client. I didn't look at B2 and didn't realize there was a client for Linux that can work with it. It looks like it comes out to $5/TB/mo roughly... which is reasonable, but still $35/mo for 7TB. That is reasonable, but i guess I was initially hoping Crashplan would work and was excited about the $5/mo for unlimited... so now i have to adjust my internal expectations.

B2 is different from their backup product from their mainline backup product. We were using the latter and then had a nasty surprise that they only retain data deleted on the client for 30d. B2 is much more "generic cloud storage" than specifically backups, it's up to you to run/manage the client side but it does give you much more freedom. It's more akin to S3/Glacier but cheaper than both and less restrictive than the latter.