Raid10, must use or another Raid limits?
-
@Jimmy9008 said in Raid10, must use or another Raid limits?:
In terms of VMs, from memory (im at home) the blades run around 380 VMs. The storage side (aka 170TB number) comes from another team, so I have to just go based on that.
So... could be anything. No retention or recovery details? RTO? RPO?
-
@scottalanmiller said in Raid10, must use or another Raid limits?:
@Jimmy9008 said in Raid10, must use or another Raid limits?:
In terms of VMs, from memory (im at home) the blades run around 380 VMs. The storage side (aka 170TB number) comes from another team, so I have to just go based on that.
So... could be anything. No retention or recovery details? RTO? RPO?
Retention: 20 days, daily incremental. RTO, RPO - nope.
-
@Jimmy9008 said in Raid10, must use or another Raid limits?:
daily incremental.
Such a meaningless thing to have included. So frustrating.
-
@Jimmy9008 said in Raid10, must use or another Raid limits?:
Retention: 20 days
That's at least pretty easy.
-
@Jimmy9008 said in Raid10, must use or another Raid limits?:
@scottalanmiller said in Raid10, must use or another Raid limits?:
@Jimmy9008 said in Raid10, must use or another Raid limits?:
@scottalanmiller said in Raid10, must use or another Raid limits?:
@Jimmy9008 said in Raid10, must use or another Raid limits?:
At the end of the retention period, it injects the oldest increment in to the original full backup... is that not possible with tape?
Definitely not possible with tape. But that's just one of the many assumptions that need to be changed. Someone did your backup planning based on false assumptions. You need to back up and decide on goals and make all decisions based on those goals.
However, what you describe isn't exactly how Veeam works. Also, Veeam is a company, not a product, there are multiple free products from them. And they all work a bit differently.
https://www.veeam.com/windows-endpoint-server-backup-free.html
That one.
Good thing that I asked, this is a case where me being pedantic really did result in the answer no one would expect. "The free version of Veeam" means this one to almost all people: https://www.veeam.com/virtual-machine-backup-solution-free.html
Does the version you put here allow scheduling with the free version, and unlimited VMs?
If free is the target point for the software that does this, while Veeam has an option, it may not be the best solution.
-
Honestly speaking, since you can't afford any licensing for this, you should only be looking at open source solutions. Not even things that are free and closed.
Just removing the "licensing overhead and conversation from the table".
-
@DustinB3403 said in Raid10, must use or another Raid limits?:
Honestly speaking, since you can't afford any licensing for this, you should only be looking at open source solutions. Not even things that are free and closed.
Just removing the "licensing overhead and conversation from the table".
If we were building with full freedom, definitely. UrBackup, Tape, we could solve the problem pretty easily from a business perspective.
But given the strict requirements, our hands are tied.
-
I would not do as what has been discussed.
I'd do 3 arrays of 8 x 8TB 3.5" SATA enterprise drives on software RAID 6. That's about 150TB. If it's not enough just go with 4 arrays instead.
Time to rebuild a 8TB drive at 100 MB/s is 8000000/100=80000s=22 hours. That's nothing.
Probability for drive failure is 4% per year for each array.If you have unrecoverable read error (probability < 9% for the above config) during rebuild it's not the end of the world. That's old hardware raid thinking. You just have to clone the offending drive, replace it and rebuild again. You will end up with a byte somewhere on the array that are incorrect but the rest of the data will be fine. Backups should have file hash (checksum) to verify it's integrity so no big deal in this particular case.
8TB Exos Enterprise SATA drives are around $250. So 24*250=$6000 for the drives.
Software raid won't need a raid controller and has very low CPU and RAM requirements on HDD arrays - if you use a reasonable file system. If you're using Supermicro you should pick a chassis with a SAS port expander, not the one mentioned earlier in the thread.I would use the arrays as individuals so you can take one of them out of commission if you like and run backups on the others instead. Maybe even put it on two servers instead of one.
-
@Pete-S the issues here is you're splitting the array. The disks all could still fail, and a resilver would need to be done anyways.
In what world if a few bits being screwy acceptable?
That means that data is useless.
-
@scottalanmiller said in Raid10, must use or another Raid limits?:
@DustinB3403 said in Raid10, must use or another Raid limits?:
Honestly speaking, since you can't afford any licensing for this, you should only be looking at open source solutions. Not even things that are free and closed.
Just removing the "licensing overhead and conversation from the table".
If we were building with full freedom, definitely. UrBackup, Tape, we could solve the problem pretty easily from a business perspective.
But given the strict requirements, our hands are tied.
The backup solution is up for design. We aren't dealing with the existing file servers.
Urbackup on the white box, with a MD raid 10.
-
@DustinB3403 said in Raid10, must use or another Raid limits?:
@Pete-S the issues here is you're splitting the array. The disks all could still fail, and a resilver would need to be done anyways.
In what world if a few bits being screwy acceptable?
That means that data is useless.
I'm not sure what you mean. But a backup is a backup, so the original is still in place. If you have 100 incremental backups and you know that number 65 has bit errors you have to schedule a full backup on the VM that the backup belongs to.
It's a numbers game.
-
@Pete-S it's not just a numbers game, that's a failure of your backup. And if you require data from that backup, then your done for.
It may work, it may not. But it's not a reasonable thing to say "eh it's our only backup it may work when we need".
-
@DustinB3403 said in Raid10, must use or another Raid limits?:
@Pete-S it's not just a numbers game, that's a failure of your backup. And if you require data from that backup, then your done for.
It may work, it may not. But it's not a reasonable thing to say "eh it's our only backup it may work when we need".
RAID10 makes no difference if that is what you mean. You can have a disc failure on rebuild and then you have nothing.
That's why I say it's a numbers game. With 4% probability of having to rebuild a drive and a 9% probability of a bit error in the rebuilding process the probability is 0.4% that you will not have a backup that is OK. Now multiply that with the probability that you will need that particular backup set during the one or two days it takes for you to identify the problem and schedule a full backup of that particular VM (which renders backup with screwy bits obsolete).
And to take it a bit further. What will happen if the backup is truly lost? There is a dollar amount attached to that problem.
-
@Pete-S raid10 males a difference in that you aren't splitting the array into multiple arrays.
Split arrays never made sense in the past, it still doesn't even here.
-
@Pete-S said in Raid10, must use or another Raid limits?:
And to take it a bit further. What will happen if the backup is truly lost? There is a dollar amount attached to that problem.
The dollar amount is what needs to be calculated. It's the entire front half of the conversation that I believe @Jimmy9008 has skipped.
Knowing what a failure costs, helps a business to know what they need to spend to protect from that outage, and if it's worth the spend.
-
@Pete-S said in Raid10, must use or another Raid limits?:
I'd do 3 arrays of 8 x 8TB 3.5" SATA enterprise drives on software RAID 6. That's about 150TB. If it's not enough just go with 4 arrays instead.
That's just a complex RAID 60. Still extremely risky, like crazy risky, Rebuild times would still be in weeks, with rebuild success rather low.
RAID 10 protects you against both the URE and nearly all drive failure risks.
And with such small arrays, he'll struggle to get to capacity. At 32 disks, he's likely way slower and way more risky than a single RAID 10.
If you are going to go this path, it's RAID 60. But it's not a risk I'd be happy explaining that I was willing to take.
-
@Pete-S said in Raid10, must use or another Raid limits?:
Time to rebuild a 8TB drive at 100 MB/s is 8000000/100=80000s=22 hours. That's nothing.
That's not reality. RAID 6 rebuild on 8TB is weeks, often months. 100MB/s is not the rebuild speed. You'd have to anticipate at least 2-3 weeks for a rebuild to happen.
-
So, disregarding the fact that this seems to be a bit of a fool's errand or a case of being set up for failure..... something like this might work from a hw perspective https://www.45drives.com/products/storinator-q30-configurations.php
I gave up on reading all of the comments after a while, but has anyone touched on the target getting bogged down with all of the simultaneous backup sessions running?
If we're talking about keeping the software side of things free, would amanda / zmanda be an appropriate solution?
-
@Pete-S said in Raid10, must use or another Raid limits?:
Software raid won't need a raid controller and has very low CPU and RAM requirements on HDD arrays - if you use a reasonable file system.
File system does not interact with the RAID array. RAID doesn't even know if it has a file system.
-
@notverypunny said in Raid10, must use or another Raid limits?:
So, disregarding the fact that this seems to be a bit of a fool's errand or a case of being set up for failure..... something like this might work from a hw perspective https://www.45drives.com/products/storinator-q30-configurations.php
Those are never meant for SMBs to use, those are RAIN cluster nodes. Incredibly risky in a RAID configuration. BB never meant for that design to be used outside of a disposable node, massive cluster setup. It's not a mistake on BB's part, it is SMBs misapplying knowledge from another field, or in the case of 45drives, a vendor preying on SMBs not knowing better. I've had to rescue companies that made this mistake with much smaller arrays.
https://smbitjournal.com/2016/10/smbs-must-stop-looking-to-backblaze-for-guidance/
