HA - Backups

One of the important things to understand about backup is the backup types.

Backup considerations

Some of the considerations we have over backup would be things like: is the data recreatable, how critical is the data, what's the cost of keeping that data, what is our RPO and RTO that we're advertising to our customers, what kind of disruption to our customers would it be if we didn't have that data, what are the copy times, what is the management time that it takes to manage that data, how much security do we need around that data, and how dynamic is it — how much change to that data is occurring.

All those considerations go into things like how often, what's the frequency that we're going to back things up at.

Backup types

Let's say this is an hour in between each one of our backups, and we're capturing these changes. Here's change one, here's change two, and three are happening at this time. Then we go a couple of times when there's not really anything being changed, and then we see another change.

One way we could do it is we just do a full backup every time, so this gets backed up wherever these iterations are, and it's not going to change — it's just always going to get backed up, whether there's a change or not. That's a full backup. The problem is that a full backup every time could be problematic, not just from a monetary standpoint of backing up everything all the time, but also from a processing standpoint and a resource standpoint. A lot of times backups can be a little bit problematic on our systems, especially when we're doing full backups. So we may want to consider one of these others, like a differential, incremental or synthetic backup.

A full backup backs up everything. A differential backs up only the changes that have happened since the last full. Let's take a look at what that looks like. Let's say we have a folder here, and the folder has lots of files in it, and we're adding more files to it — we're adding file one, file two, file three, file four, and so on throughout the day. We're going to start the day by doing a full backup of the full folder, so the full folder gets backed up at the beginning. From here we're going to do a few differentials, and then midday we're going to do another full, and then we're going to do some more differentials.

At 9:00 a.m. only one file has changed, so it's going to back up the changes for that file, so file one gets backed up. At 10 a.m. we see that there are now three files that need to be backed up: 1, 2 and 3. At 11 a.m. nothing has changed, so it's still going to back up files 1, 2 and 3. We do another full backup, which is the whole folder, so it gets fully backed up there again. Then at 1:00 p.m. it's going to back up whatever has changed — files four and five changed, so that'll back up 4 and 5. At 2:00 p.m. it's going to do 4, 5 and 6; at 3:00 p.m., 4 through 7; and at 4:00 p.m. it'll do 4 through 8. Then maybe we do a full backup at the end of the day again, and then the process restarts.

An incremental backup will take a backup of just the changes that have happened since the last backup — not from the last full backup, but just from the last backup. Let's take a look at what that looks like. We have a folder here, and we're going to do a full at the beginning of the day, and we'll also do one midday right here. At 9:00 a.m. it's going to do a backup of the changes, so that's going to be file one. At 10:00 a.m. we do another backup, but it's going to be a backup of the changes since the last backup, which is going to be 2 and 3. At 11:00 a.m. it doesn't have to do a backup at all, because nothing has changed. At 12:00 p.m. it backs up the full folder again. At 1 p.m. it's going to back up files 4 and 5; at 2 p.m., file 6; at 3 p.m., 7; and at 4 p.m., file 8, because those are the changes that happened.

One thing that's great about this is that it uses up less disk space, because we're backing up less information — it's not all the information since the last full. So this really helps with speeding up this process and being able to save disk space.

If incremental has all these advantages over the differential, why not just use incremental every time? The answer comes into play when we restore the data, because how many files are we going to have to restore? We did those fulls at the 8 and the 12:00 p.m. Let's say there's an issue that happens where our files get broken right here, and we need to restore to 3:00 p.m. right here. To do this, what we'd have to do is the last full, which would be at 12 p.m. right here, and then for incremental we'd have to do this file right here because we have to do four and five, and this file right here because we have to do six, and this file right here because we have to do seven. So we have to restore four different backup files here. Versus differential, we would just have to do the full, and we would have to do this one right here, which would include four, five, six and seven. So we'd have to do a lot less restores with the differential.

A synthetic full takes a little different approach. Let's say we do a full at 8:00 a.m., where we back up the whole folder, all the files are in there, and then we do increments throughout the day — on an hourly basis we're going to do increments — and then we're going to roll this all up into a synthetic full. So we've got this synthetic full at the end of the day. What's going to happen is that instead of recapturing this whole folder and all the data again, it's just going to take this backup right here with all of the increments throughout the day and combine them to create a new full. It's actually not going to go back and take from the source; it's going to take from the backups to create a synthetic full backup.

Frequency and backup plans

Something we've toyed around with this whole time is the frequency with which we take these backups. Do we want to continually take this on an ongoing basis? Do we want to do it every 15 minutes, every hour, every day, every week, every month, every year? What is the frequency with which we take this backup?

What would be an example schedule? This is one that I've commonly seen out there, and it really depends a lot on all of the variables that we've already mentioned. One thing that we want is a full backup, and ideally we'd do that at least once a day. A full backup would be great, but the problem is that when you're dealing with large amounts of data it's really quite hard to get a full backup in. In fact, at a lot of the places that I worked it was hard to get a full backup in over the weekend: you would start a full backup on a Friday and then it would continue through Saturday and possibly even into Sunday, and you want a little bit of wiggle room with that as well. If you missed that full backup it would cause problems that whole next week with the whole system. So we capture the full backup over the weekend, when things are less busy. Obviously this is going to depend on when your customers or your users are accessing this data, but a lot of businesses are off over the weekend, so that's a good time to do a full.

Then perhaps what we do is differentials daily, every day, and this is going to take a lot less time because there are a lot fewer changes. Then throughout the day, maybe we go on 15 minute increments — let's say that's our recovery point objective, so we do 15 minute increments — and so we're capturing this data all throughout that day.

So if I have to recover something that's part of this day right here — let's say something goes awry and I need to restore back to this point right here — what I would end up doing is the last full, which is going to take a while to restore, because it's taking that whole weekend just to capture it, so if I've got to restore it, that could be problematic in itself. But let's say we restore the full, and then I'd have to do the differential from the night before, and then I would do all of the incrementals from that day to get to this point right there, and now I'm back up and running to this point.