Removable Disk Backup Slow? Here are common reasons for slow removable disk backup in a bullet list. Scroll down or click to page jump to detailed explanations and suggestions.
- You Forgot to turn off Your Anti-virus software during backup.
- Your drive(s) are getting full, starting to use the inner cylinders.
- You are using small hard drives.
- Your RAID array has a failed member.
- Someone or something is performing another backup or heavy disk I/O during your backup.
- The drive you’re backing up uses a slow RAID controller.
- Your drive(s) are fragmented.
- You are using slow backup or copy software.
- You are backing up active DFS connections, Active Directory, Exchange etc with slow agents.
- You are backing up over a slow LAN connection.
- You think you’re using USB3 but it’s really USB2 or slower.
- You’ve installed multiple backup programs or programs with file I/O shims.
- You have lots of small files and folders.
- You are getting soft errors on your drive(s) because they’re failing.
- You are doing Full backups every night.
- You have low RAM.
- Another process or virtual machine is taking CPU power.
- Compression or encryption is slowing you down.
- You have a virus.
- You have Indexing turned on.
- Close Microsoft Perfmon.
- Turn off Remote Differential Compression for LAN Backup.
- Use a faster controller (or bus).
You Forgot to turn off Your Anti-virus software during backup.
Often people are running real time anti-virus or spyware scanners. These dramatically slow down performance because each read is run through the Anti-Virus I/O sub-system. Writes to the removable disk may also go through an AV scan. . Try temporarily disabling any Anti-Virus or anti-spyware software and testing the speed that way. You may need to turn off “real-time” scanning on the server and just do nightly drive scans (make sure to schedule it at a time well away from backup window). You may want to permanently disable anti-virus scanning of the removable disk.
Your drive(s) are getting full, starting to use the inner cylinders.
Did you know that the outer tracks or cylinders on a hard drive fill up first? Think of a playground merry-go-round. If you stand at the edge while it rotates things seem to be moving very quickly. If you move in toward the axis of spin (spindle) the linear velocity slows down. Similarly, data written to the outside tracks write up to 50% faster than inner tracks because bits pass under the heads at a higher linear velocity. Which means as your drive fills it will slow down.
You are using small hard drives.
Big drives with high density pack more bits into a smaller area. So given a smaller 250GB drive spinning at 7200 RPM and a bigger 3TB drive spinning at the same speed, data written (especially to outer tracks) to the big drive will lay down faster because more bits pass under the head in a given rotation of the disk. Some gamers even “short stroke” big and cheap 2TB and 3TB hard drives by partitioning them down to only a few hundred Gigs. Done correctly, this insures the data goes to the very fast, very dense outer tracks. Of course, you lose a bunch of the disk space that you paid for. But this extreme speedup technique demonstrates the concept of using dense drives for more speed.
Your RAID array has a failed member.
RAID 5, which are usually 3 or more disks with redundant data striped across them will become horribly slow if one disk fails. Since the machine stays up, sometimes users are unaware that a drive has failed except by noticing a slow down. Note that the slowdown could be either the server’s main array or, if you’re using one of our RAID 5 products could be the destination backup drives. We recommend monitoring RAID health and having an email sent out if a member goes offline. Use HW RAID Manager where applicable to notify you of events via email.
Someone or something is performing another backup or heavy disk I/O during your backup.
Hard drives don’t do well with multiple jobs causing the disk to “thrash” (seek back and forth all over the drive). If two backup jobs overlap, or if someone is running a report that requires heavy disk usage, your backup will slow to a crawl. Check Windows Task Manager to verify no other processes are running significant cpu or file I/O during a backup.
The drive you’re backing up uses a slow RAID controller.
Price pressure has forced server vendors to create small business servers with anemic RAID controllers built-in. Resellers who are aware of the difference will often upgrade the RAID controller for one with more processing power an RAM. In theory it should be faster to read from multiple drives but we’ve seen several examples where RAID performance is slower than a stand alone drive. Upgrades to RAID controllers can sometimes be done by adding RAM or processor power. Run our programs called “Fakeback” and TRMark to determine if your source drives are running slowly.
Your drive(s) are fragmented.
Most of us have heard about defragmenting our hard drives. When using removable disk backup, you have TWO potential fragmentation problems: The source drive and the destination drive. Both of these can potentially fragment over time. If the Removable disk has been heavily used, and/or already contains existing data, you may achieve some advantage by defragging it as well as your source drive, but it may not be necessary if your backups delete or overwrite the drives (full backups). Microsoft includes defrag software with your server OS but you may want to invest in a program that does a better job, including putting everything to the outer tracks in contiguous order. Be aware that running defrag on a hard drive that is doing incremental backups may “break” the incremental scheme causing the software to have to do a full backup after every defrag. Try defragging (preferably with a 3rd party defragger) the source drive and read http://www.ntfs.com/ntfs_optimization.htm for more tips. For example, you might want to use larger cluster sizes on your High-Rely removable disks since backup tends to write data in large blocks. You may also want to look at a product called Ultimate Defrag by Disktrix. Do not defrag SSDs, as this can shorten their life.
You are using slow backup or copy software.
We like to run real world speed tests with block level imaging products like StorageCraft’s Shadowprotect. File backup is usually slower than block level imaging. Any option to verify the backup during performance testing doubles the backup time and should NOT be counted in the benchmark, although you may need to consider it for the entire backup window. It IS a good idea to perform some sort of verification via CRC checking, data comparison, or test restore. Try our benchmarking tools Fakeback and TRMark or something like CrystaldiskMark to get a sense what speeds you should be seeing. You should also be aware whether the backup program is using buffered or unbuffered I/O. Unbuffered I/O (or a raw file copy) is preferred when attempting to copy a large file from one location to another. To test, try using Xcopy /J or eseutil to copy file(s) using unbuffered I/O (ESEutil is a program that comes with Exchange) as described here. As an experiment, you might also try Microsoft’s free RichCopy, which supports multiple threads. We are not sure if it supports unbuffered I/O.
You are backing up active DFS connections, Active Directory, Exchange etc with slow agents.
Some backup “agents” such as Exchange, open file, or SQL agents, may backup much more slowly than native file backup.
You are backing up over a slow LAN connection
Direct Attached Storage (DAS) is generally faster than NAS (Network Attached Storage). We sell both, and each has it’s place. For maximum speed use DAS over eSATA or USB3. Backup speeds taking data off remote servers over the 100MB Ethernet network will be slower than Gigabit Ethernet. Standard 1500 byte frames are slower than jumbo frames (9000 plus bytes). Some speed increases might be achieved using multiple ethernet ports. This is known as Link aggregation, bonding, or NIC teaming. In order for this to work your ethernet switch needs to support IEEE 802.1ax Link Aggregation Control Protocol (LACP) or you’d need to do dual hardwired connections from the NAS to the server (dedicated ethernet links for backup). Most experts say doubling your ethernet ports doesn’t double your speed, and that performance increases are modest. LACP requires the Ethernet NIC drivers to support it, although it is rumored that in Windows 8 server link bonding will be done at a higher level in the operating system, allowing more dissimilar ethernet cards to be bonded.
You think you’re using USB3 but it’s really USB2 or slower.
If the host subsystem is inadvertently using slower USB (1.0, 1.2, or 2.0) versus USB3.0 , it will make a huge difference in performance. Expect 25-35 Mbps (100 Gigabytes per hour) on USB 2.0. Expect 200-400 Gigabytes per hour on eSATA or USB 3.0 if the other problems in this list don’t slow you down. Use TRMark or BurnIn Test Pro to get quick numbers.
You’ve installed multiple backup programs or programs with file I/O shims.
The same comment we made in #1 about anti-virus scanners is true for *any* piece of software that hooks into the I/O. Any program that installs itself into the operating system read/write stack (I/O shims) can dramatically affect read and write performance. This can include other backup programs. Be very careful about installing 2 or more backup programs on a server because often both are looking at all file I/O to determine what changed to help with incremental nightly backups. This can lead to conflicts. Multiple backup programs can break Windows open file (VSS snapshot) systems. For example Acronis and Storagecraft have problems when installed on the same system.
You have lots of small files and folders.
Plays in heavily to performance. Larger files and fewer deep directories will be much faster to backup than lots of smaller files with complicated directory structure. It’s hard to do much about this but if there are lots of small files that could be the problem. Imaging software that reads at the block level may be better for lots of small files (Acronis True Image, Symantec System Recovery, StorageCraft ShadowProtect, Appasure, DoubleTake, newer Windows native backup).
You are getting soft errors on your drive(s) because they’re failing.
Errors on the source or destination hard drives will slow performance. Modern hard drives will attempt to correct “soft errors” by retrying. Drives that are slowly failing may perform failure prior to a complete failure.
You are doing Full backups every night.
Incremental backups move less data and can be more efficient. Using software that will do “synthetic backup” or roll the incremental backups into a single full can improve comfort levels with an incremental back up scheme. You can also eliminate things like Page files, Hibernation files, Temp directories, Recycle bin, and other things non-critical to your backup
You have low RAM
Systems with very low amounts of available RAM will use the pagefile excessively, dramatically slowing performance.
Another process or virtual machine is taking CPU power.
If the CPU is heavily loaded, doing any other type of task, that will obviously affect backup performance. For example, if a server were being heavily used for database access, running spyware, or doing computations during the time of the backup, then fewer CPU cycles would be available to the backup process. Do a CTRL-ALT-DEL, and on the processes tab, arrange processes by “CPU” to see if there is a process taking an inordinate amount of CPU during the backup. Try not to install programs or services that stay in memory on a server. Software bloat kills performance.
Compression or encryption is slowing you down.
If compression or encryption is turned on, either on the source NTFS disk, on the destination disk, or in real time while using the backup software, it can slow down backups. Oddly sometimes it speeds it up if compression is done before the data is transferred to the disk because with fast processors that compress quickly, less data is moved to the destination drive. Don’t be surprised if you see either faster or slower backup speed with compression turned on/off as results will depend on the environment (hardware vs software compression and amount of CPU horsepower you have) and the nature of the files (how compressible they are). Some already compressed files actually expand slightly if compressed again, resulting in more space taken on the backup media and slower speeds.
If your disks are vibrating, you may be having hundreds or thousands of disk errors that disappear upon a re-read. This will manifest as extremely slow performance. The vibration could be other drives in close proximity, fans, or other mechanical devices. We’ve seen demonstrations where a benchmark is running on the hard drive, someone slaps the side of the rack, and performance drops for several seconds.
You Have a Virus.
It almost goes without saying that many spyware, rootkit, or virus infection will cause performance problems. Spyware can kill performance!
You have Indexing turned on.
Indexing Service (originally called Index server) was a Windows service that maintained an index of most of the files on a computer to improve searching performance on PCs and corporate computer networks. It updated indexes without user intervention. In Windows 7, it has been replaced by Windows Search. If your server is indexing either source or destination drives, turn it off. Indexing Service is still included with Windows Server 2008 but is not installed or running by default.
Some users have reported that running Perfmon has caused slow disk I/O.
Turn off Remote Differential Compression for LAN Backup.
On Vista or higher Windows some users have reported slower file transfer when RDC is turned on. Although Microsoft recommends against turning it off, you might experiment if you’re having speed problems. To turn it off go in Control Panel / Programs and features / Turn on or turn off Windows features and uncheck “Remote Differential Compression”.
Use a Faster Controller (or Bus)
Whether you’re using USB3 or eSATA, there is a speed difference between bus types (PCI-X, PCI, PCI-e), widths, and brands. For example you may see speed differences between Renesys based USB3 chipsets and TI chipsets and a controller with 1x connector may be slower than 4x (4 lane PCI-e). Additionally the version of the bus makes a difference because, for example PCI-e 1.0 is slower than PCI-e 2.0. The motherboard bus connectors must support the full speed of the controller to take advantage of later generation speed increases. Only when they are well matched and the driver is optimized will the highest speeds be attained.
This list is not comprehensive but covers most of what we know about backup performance issues. Here is a similar document from Symantec if you want to get another written perspective on common speed issues.