Sluggish Copy speed from one disk to another

[bence8810]

Hi,

I have a problem on 2 HP servers with copying. I am collecting backups on them throughout the week, and on Monday, I copy the whole backup to an Externam HDD. The copy speed is good, and it takes less than 2 hours to copy 150GB to the external DISK. When this is done, I copy everything to the 2nd internal disk, and delete it from the 1st one, so I can start collecting network backups again.

From the 1st disk to the 2nd, it takes 7-8 hours for 150GB of data. What could be the problem? Its quicker to actually copy it to an external drive and then back to the 2nd disk.

The two servers are HP ML110 and ML150. Both have 500GB SATA disks, 2 each, non mirrored, standalone.

Thanks for any help,
Ben

 

[Mmitchell57]

First thing that would come to my mind is the interfaces. When you say you are copying to an external hard drive, the first thing that comes to mind is USB. It's possible you might be going from usb1 to usb2 and it's going insanely slow because of that.

Just an idea since there's not many details.

 

[bence8810]

Hi

Sorry if I was vague.

The External HDD connects via USB2, and the other 2 HDDs are internal SATA drives, connected to the SATA controller that is also internal to the server.

To summarize things:

There are 4HDDs inside each server. 2 are mirrored for the OS, and ther other 2 are 500GB drives on their own, but on the same internal Controller. Lets call the 2 internal HDDs A and B, and the USB External drive C.

A collects all the backups, and I copy the weekly backups to C on Monday, then copy them also to B and delete from A.

B keeps it for an extra week, should there be anything needed for a restore.

Copying from A to C takes 2 hours for 150GB, the same amount takes 8 hours copying from A to B. A and B are both on the same controller, which is an Adaptech Serial ATA Hostraid.

Thanks,

Ben

 

[FredWagner]

It sounds like the problem is caused by the same SATA controller handling both the cource and destination drives when you're copying from A to B. With SCSI in days of old, this wasn't a problem, because most of the smarts were in the drives. with SATA, I think most of the smarts are in the controller, and it's having to play catch with itself. If you and add another SATA controller card, it might help. If you could go with SCSI drives, that would help for sure. Two fast SCSI drives on the same SCSI bus would copy really fast....

Fred Wagner

 

[58sniper]

I agree. SATA drives are painfully slow (7200 RPM, usually). SCSI drives are typically 10k RPM or 15k RPM. Combine that with the SATA controller handling the traffic for everything. The advantage to SCSI controllers as well is that some had memory right on them.

Also check your AV solution. Is it doing a check on the file that's being read and the file that's being written?

Pat Richard, MCSE MCSA:Messaging CNA
Microsoft Exchange MVP

 

[bence8810]

Hi

Thanks for the posts. I do not have an AV solution, that was the first thing I made sure, besides wouldnt that also check data going from SATA to USB?

I have other older machines with IDE drives, 5400RPM, connected to good old motherboard IDE controllers, and they copy much much faster also.

As for the ML150, I just cannot believe it would be so slow, specially that it has the SATA hotplug RAID system, which leaves no space for putting other drives in, only the 6 drives in its bay. Would such a system be slow like this by default? Otherwise its not a slow machine, for example when I get data through the LAN, it copies it also much much faster. Maybe 3-4 hours for the 150GB, or even less.

So you are suggesting this will never improve, unless I drop SATA, or for example put a silly IDE drive hanging on the cable or sitting on the bottom of the case, and copy things from SATA to regular maiboard IDE? This would then speed things up?

I was much more thinking about a mistake I must have made with the RAID card, etc, which slows things down.

Thanks again for all input, do you have any suggestions? SCSI is not possible, this is just a Backup server, so it must remain low-cost. All other servers are SCSI 15K RAID5 to take the office load, but this server just stores data which will most likely not be needed ever.

Ben

 

[bence8810]

Hi

Still searching for the solution. Actually I need to correct myself above, and this may change the whole situation.

I tried again copying from the USB to the HDDs, and it is also slow. So basically copying from the drive to an outside source to fast, but to copy to the drive is slow. No matter if I copy from another HDD or from external HDD, its always slow to write to the disk, but fast to read from the disk.

Would this be a driver error? What should I check?

Thanks

Ben

 

[FredWagner]

Could it be that the drive is badly fragmented ? See if DEFRAG will run on it. The situation might be that when reading from the SATA drive, the controller is caching the files, but when writing, it has to keep looking for the next open blocks. Can you specify Write-caching on your controller ? Do you have Virus protection running in this backup server? might it be scanning the files as you write them ?

Fred Wagner

 

[bence8810]

Hi

Thanks for the pointer, I see what you mean by the Caching. I could only get to it today, so I went ahead to the Server room, rebooted the machine, and saw that Write Caching was disabled on all drives. I have 4 drives, 2 in mirror holding the OS, and 2 500GBs standalone.

I enabled caching on the RAID and on the 2 drives, rebooted, and the RAID BIOS started giving me Warnings, that Caching is risky, etc. I quickly went back, and disabled Caching on the RAID and left it on for the 2 standalone drives. Rebooted, still gave me warnings, but I quickly ignored it, and now I am copying the Files. Will report back if the Speed was increased.

Just to verify, what I did is what you had in mind?

Defrag is not an issue, the drive is perfectly healthy, almost brand new, and all files are consistent.

Thanks

Ben

 

[bence8810]

Hi

No, it didnt help. The copy speed is still very very sluggish, takes 7-8 hours for the 200GB or so. Again, copying from one HDD to another. I now have Write Cache enabled, but still realy slow. Anything else I could try? This cripples my backup capabilities, as I would like to keep the backups for 2-3 weeks, but for that I need to copy it / move it from one HDD to another.

Maybe I should try to copy it with Xcopy? Would that be better/faster?

Thanks

Ben

 

[FredWagner]

Are you copying by Drag and Drop ? That's very slow, because it gets the GUI involved. Try copying with a batch file, and use XCOPY or XCOPY32, whichever is active on that system. You can launch the batch file from the GUI, and be sure to put a pause at the end, so you can see what it did. You might even set it up to give a timestamp at the beginning and the end, and pipe the results to a .txt file you can review as a log.
Internal drive to drive copying should go much faster than what you're seeing. I have a server with about 300GB on it, that has a 100BaseT connetion, and it takes over 13 hours for a full backup across the network. I'm due to replace the server (it's a dual Pentium 700, in service for over 6 years), with a new one with a Gigabit NIC.
any drive that gets different files copied to it can get fragmented. The drives you're copying FROM may be badly fragmented. It they're not too large (over 200GB ), you should be able to defragment them. that could make all the difference. Try it!
BTW - what else is the server doing ? check Task manager for the processes table and Performancegraph and see what's going on during the copy process....

Fred Wagner

 

[bence8810]

Hi

Well, now I tried to copy one file with drag and drop and one file with xcopy. Guess what. Drag and drop came out on top.

You put the bug in my head regarding the defrag, even though the volume has only 10 files or so, but definitely less than 15. I ran a defrag analysis on it, and it said You should defragement your volume. Wow, was I blown out of my theory in an instance.... I thought defragmentation is only necessary on volumes with lots of small files.

Long story short, I cannot make the defrag now as the new backups are being copied to the volume starting in less than 20 minutes, so its impossible to complete before that.

You say volumes less than 200GB are fine for defrag, but my disks are 500GB and one partition. Is that bad? Can I still defragement it?

Thanks

Ben

 

[FredWagner]

Ben -
I don't know the official reference, but in practical experience, on the big old server I am responsible for, a 2001 model IBM Netfinity 6000R, with a 28 drive RAID array, I can defrag the 16 GB OS volume, I can defrag the 135 GB SQL volume, but when I try to defrag the 888GB data volume, the message comes up that "The Volume Data (E has 64kb per cluster. Disk Defragmenter cannot defrag NTFS drives that have more than 4K per cluster."
Maybe there's a product out there that will defrag volumes with larger cluster sizes, or maybe it's improved in Win2K3. I have Win2K3 on some other servers, but they have much smaller drives...
If defragging works and helps, do it fairly often! Back when I ran Netware servers, they never needed defragging, because their Disk I/O was written to use 'elevator seeking' and it would pick up clusters for files requested in smooth passes across the disk. Not sure why Microsoft couldn't do the same thing, but that's how it seems to be!

Fred Wagner

 

[Bigdaddynavarre]

You may want to consider Diskeeper defrag software and using a backup software like Yosemite with its Bare Metal Restore capabilities. Do the sata controller and sata disks support native command queing? That may affect your speed.

Keep the thread going with your results!

 

[FredWagner]

Ben -
Your comparison of copying with Drag and Drop vs copying from the command prompt may have been compromised - which one did you do first ? That fist copy operation would have put the file in read cache, so whatever copied it next would have been reading from cache, rather than from the disk itself. Once it's in cache, you're testing the speed of the Memory access and the bus or network connection.
Do keep us posted on your results!

Fred Wagner

 

[YoRafael]

I have the same problem. It's very slow copying files and slows down the server and everything you are doing with it. I'm thinking to disable the RAID option to see if it will help.

 

[cpjust]

You could try running HD Tune:

http://www.hdtune.com/

and check the Transfer speeds of each disk. Although I wouldn't be surprised if it wasn't just a fragmentation problem. You should schedule Defrag to run (after hours) on a weekly or monthly basis.

Regarding the SATA controller & drives. If I read correctly, you have 4 drives (2 of them RAIDed) connected to 1 controller, is that right?
Do you have any empty PCI slots on the motherboard? You could add a 2nd SATA controller and move 1 or 2 of the drives over to the other card (make sure they're on different channels/cables too).

 

[bence8810]

Hi

I will try this HD Tune and see what its showing for speeds of the disks.

As for now, I have disabled the Write Caching completely, as my Monday backup couldnt complete, it crashed halfway spilling errors to the event log every 2 seconds. The whole event log (has a size setting of max 500kb) was covering 10 minutes or so, as it was full after that.

I rebooted, now it seems to be back to normal with the slow speeds, but at least reliable.

And yes, your assumption is right, I have all 4 drives on the same controller, 2 of them are RADIed and two are standalone. However I dont understand why would a RAID controller come with 6 SATA HotPlug slots if it wasnt working properly. Of course as a last resort, I will buy a 3rd party SATA controller, and do it with that, but HP will take a dive in my eyes if it has to come to that.

Again, thanks a lot for your time, I will report back after I try the hdtune

Cheers

Ben

 

[FredWagner]

If there is any way you can finance it, consider migrating to SCSI drives. with SCSI, the individual drive controller has a lot of intelligence and does most of the work, just uses the bus for passing data. It sounds like the SATA drives (I have NONE, all of my production servers with RAID are SCSI), are like IDE, just a pass-through interface, with the main CPU doing all the work. You try to copy from one drive to another, it will soak up all your CPU cycles.
I once inherited a server with IDE drives that had been set up for RAID0. One of the drives failed, and the server because unusable because the CPU was saturated trying to rebuild the failed drive. I salvaged the situation by disconnected the bad drive, the got a new server built, and coped the data. The new server was SCSI, of course!
SATA and USB are probably superb solutions for home systems, but in my opinion, should not be used in enterprise production environments with RAID of any flavor.

Fred Wagner