Windows "compressed" drives and backup performance

[BonzaiDuck]

I could probably post this on the "OS" forum, but it probably belongs here.

The scenario: From my thread on the "OS" forum and possibly other threads here and there, you might know that I have been belatedly building a replacement home-server from some surplus hardware, Hitachi enterprise disks and a SuperMicro 8-port controller with Win Server 2012 R2 Essentials. So far, it has been an interesting project, although some have told me to abandon 2012 R2 and just use Windows 10. I'm not yet ready to do that.

Now . . . . for backup, Win Server Backup (for the server) doesn't work according to my satisfaction. I've looked into AOMEI Server Backupper and Macrium Reflect Server versions -- both priced at $170 to $275. I am not inclined at this point to shell out for either of those softwares. Instead, I've set up a $40 license to Syncback SE from 2BrightSparks to make a week's cycle of full plus incremental backups, and it's working fine as far as I can tell.

The target backup drive is a 2.5" 4TB Seagate Barracuda. I believe it is currently connected to one of the SATA II ports on the motherboard, with both of the SATA III mobo ports occupied for other uses. I'm thinking of moving the drive to the SuperMicro controller on a fifth SATA connection from the second SFF-8087 forward breakout cable. But there shouldn't be much loss (if any) to performance of this drive as currently configured on an SATA II port. Most electro-mechanical disks still have throughputs well within the SATA II limitation.

The source drive of the data is a virtual drive of my Stablebit drivepool -- all drives connected to the SATA III SuperMicro ports, and as far as I can tell, with throughput at or above 150 MB/s. Gigabit Ethernet is still going to cap performance at that level. But the backup represents internal transfer from the SuperMicro controller to the onboard Intel SATA controller.

The target Barracuda drive was set up to use Windows compression to save space. The initial "Full" backups proceed slowly, and I see throughputs between maybe 6 MB/s and 60 MB/s during the backup process.

Is the Windows compression feature slowing down the backup process? Would it make any difference to move the backup target disk to the PCIE SuperMicro controller and an SATA III port?

What's your opinion?

 

[Ketchup]

Well, it depends. If your CPU is already running a good bit of load, you might see a difference. However, since the files being backed up would actually be smaller, it may actually be faster, as long as your CPU has plenty of room left to handle it.

 

[BonzaiDuck]

Well, it depends. If your CPU is already running a good bit of load, you might see a difference. However, since the files being backed up would actually be smaller, it may actually be faster, as long as your CPU has plenty of room left to handle it.

Well, there's no real load on the CPU. It shows about 8% to 15% at any given point. The i5-3470 mostly runs at 1600Mhz [EIST lowest], and might bounce up to 2300 or 2600 here and there.

I can't help thinking that the SuperMicro Marvell controller uses CPU PCIE lanes, while the backup data Barracuda connects to the board's SATA ports and the chipset.

Even so, with incremental backups, it wouldn't matter as much, with many files being the same as they ever were initially.

 

[XavierMace]

The target Barracuda drive was set up to use Windows compression to save space. The initial "Full" backups proceed slowly, and I see throughputs between maybe 6 MB/s and 60 MB/s during the backup process.

You need to keep in mind when you say your drives are good for 150MB/s, that's with large (meaning sequential) file transfers. You're not going to get anywhere close to that when transferring large quantities of small files (in other words random I/O). For example at work, I've got a server that contains archived configs for all the network devices we manage. So you're talking around a million files all of which are only a couple of KB in size. That took 2.5 days to copy to another server.

If you benchmark your drives what do their random 4k speeds look like? That's more reflective of what you're doing.

 

[BonzaiDuck]

You need to keep in mind when you say your drives are good for 150MB/s, that's with large (meaning sequential) file transfers. You're not going to get anywhere close to that when transferring large quantities of small files (in other words random I/O). For example at work, I've got a server that contains archived configs for all the network devices we manage. So you're talking around a million files all of which are only a couple of KB in size. That took 2.5 days to copy to another server.

If you benchmark your drives what do their random 4k speeds look like? That's more reflective of what you're doing.

Well, those are good points, and I'm always looking for other opinions about it. I was monitoring the throughput of the virtual drive of the pool, and I thought that here or there I saw up to 250 MB/s, but I'd have to look again. 150MB/s sounds "more like it," which was still the maximum sustained throughput in most newer HDD specs. But you're correct about the small files, and I didn't yet check with a benchmark tool. The performance of the 2.5" Barracuda looked more like what I'd seen on an old HDD in a USB2 external drive box, but then that external drive didn't come close to the ~60MB/s numbers at the high end for the barracuda.

The fact that it took two days to copy your million files to another server is reassuring that there's nothing likely wrong with my system or the disk. It all checks out; I ran the diagnostics; yesterday, I defragged the backup disk.

I'm still evaluating the use of 2.5" consumer laptop drives in desktop systems. The Barracuda Pro drives only come in the 3.5" footprint. This Barracuda seems more like a miniature 3.5" HDD than a flimsy laptop disk. It's 15mm thick. And -- of course -- it spins at only 5,400 RPM.

I think I'm "good to go" with this. The system is jammed with Hitachi and Seagate drives and one of the 2.5" bays is empty. Plenty of SATA ports left, and I was thinking of adding an SATA SSD cache. But the existing speed of disk access is capped by the gigabit Ethernet speed, so it's probably not worthwhile.