Just a quick one.

 

I use CCC to backup three disks onto two 4TB disks in my server.

 

I typically had all the disks going to say /mnt/disk1/Backups/Disk A

I changed it to /mnt/user/Backups/Disk A when I did some changes a few months back.

 

Since then performance has been terrible - for example a disk that took ~20 mins to backup 1.8TB of photo files would now take 5+hours.

Tonight I reset the folder path directly to a disk - performance back to 20 mins!

 

It seems that the overhead on user shares completely overwhelms the cached write advantages?

 

Anyone else seen anything similar? (This is with 6.1.9 and 6.2rc2)

Since you didn't follow any of the rules for a defect report, I have moved this to General Support.

Sorry must have had a moment - was supposed to be in here

What is CCC btw?

 

I have got no perceivable diff in term of performane. My guess is possibly the backup app is trying to write multiple files at the same time to the same disk, leading to some random access penalty. Somehow by pointing to a specific disk, it doesn't do that?

 

Or your disk1 is just faster than the rest of the array.

Technically, writes to user shares are slower than writes to disk shares.  But would you even notice the difference?  Very unlikely.

 

On the other hand, if one of your disks is currently disabled and writing to the user share is winding up writing to that disabled disk, then writes will be significantly slower than directly to a disk which is not disabled (or if the user share system chooses a disk which is not disabled)

 

Additionally, depending upon your split levels, writes to user shares could result in one file being written to one disk, the next file being written to a different disk (and having to spin up the drive) which would impact your throughput.

CCC is Carbon Copy Cloner - it uses rsync as a base and uses that to enable perfect backups along with history.

 

I have three configs, each doing a disk to a folder on either disk1 or disk2.

 

So the following takes about an hour to complete, all running at the same time. (about 3TB, no idea how many files).

 

Mac HD 1 -> /mnt/disk1/Backups/Mac HD 1

Mac HD 2 -> /mnt/disk1/Backups/Mac HD 2

Mac HD 3 -> /mnt/disk2/Backups/Mac HD 3

Mac HD 4 -> /mnt/disk2/Backups/Mac HD 4

 

Originally I had it as:

Mac HD 1 -> /mnt/user/Backups/Mac HD 1

Mac HD 2 -> /mnt/user/Backups/Mac HD 2

Mac HD 3 -> /mnt/user/Backups/Mac HD 3

Mac HD 4 -> /mnt/user/Backups/Mac HD 4

 

Which took 5+ hours.

 

Something here is surely very wrong!?

 

System is as far as I can running perfectly, no errors in the syslog, cache drive is doing 500+MB/sec read/write (SSD), all disks are healthy.

 

Looking at top during the backup the user share process is using 50%+ cpu.

Suspicion is all you 4 tasks try to access the same disk at the same time at various points, leading to slowdown due to seek time.

By splitting each task to a specific disk, you ensure at worst there are 2 tasks accessing the same disk at the same time.

 

That's 2x improvement in speed and likely more as seeking lag isn't exactly a linear thingie.

A diagnostic covering the time of the activity might shed some light.

Suspicion is all you 4 tasks try to access the same disk at the same time at various points, leading to slowdown due to seek time.

By splitting each task to a specific disk, you ensure at worst there are 2 tasks accessing the same disk at the same time.

 

That's 2x improvement in speed and likely more as seeking lag isn't exactly a linear thingie.

 

The folders arnt split over disks? One disk goes to one disk on the server. Why would a backup for HD 4 be seeking for files on disk 1?

 

 

 

 

A diagnostic covering the time of the activity might shed some light.

 

I'll change the config when I get home and try - last time I looked though it was nothing out of the ordinary.

It depends on how you set up folder split level. Depending on the configuration, it is entirely possible that Mac HD4 folder is being split across both physical disks. Or Mac HD1-HD4 are all on the same disk. Or combination.

 

You can ssh in and use MC to see the exact distribution.

 

My point was basically if you are running tasks in parallel, you have to be mindful of distribution.

It depends on how you set up folder split level. Depending on the configuration, it is entirely possible that Mac HD4 folder is being split across both physical disks. Or Mac HD1-HD4 are all on the same disk. Or combination.

 

You can ssh in and use MC to see the exact distribution.

 

My point was basically if you are running tasks in parallel, you have to be mindful of distribution.

 

 

I see what you mean, but no the split level is set such that folders under the main 'Backups' folder is not split.

 

I didn't want files from one disk split over two disks - completely pointless for my setup.

Have u checked the exact distribution using MC?

Have u checked the exact distribution using MC?

 

Checked via SSH - Disks 1 and 2 are on array disk 1, disks 3 and 4 are on array disk 2.

 

It is actually quite a bit of faff to put it all back and my write speeds are 55MB/sec anyway so I'll just leave it as is.