7

One of my customers is planning to set up a new document store, which will probably grow by 1-2TB per year. One of my co-workers says that Windows performance is extremely bad if it has a single NTFS volume that is bigger than five or six TB. He thinks that we need to set up their system with multiple volumes so that no single volume will exceed that limit.

Is this a real problem? Does Windows or NTFS slow down when the volume size reaches several terabytes? Or is it possible to create a single volume of 10 or more TB?

3 Answers 3

9

I've made 30TB volumes before. They were holding large files, so that greatly assists in avoiding performance degradation. No problems there versus smaller volumes.

Where problems might begin to occur is if that large filesystem builds up enough files and directories to get to the insanely big ranges. I'm talking 20 million files and directories or more levels. At that point the MFT is likely to be pretty fragmented and on RAM constrained servers that might start to introduce some directory-display performance issues. Actual access to files should not be affected, just fileOpen() and directory scanning operations.

Is it a real possibility? Yes, but the effect also holds true for similarly large filesystems on other platforms and for the same reason. Also, the operations most impacted by the performance drop may not even be the ones you're worried about. Since this sounds like a new setup, it doesn't sound like you'll be using this 10TB on a server with only 1GB of RAM so you shouldn't have to worry.

Where the worry about size came from is likely WinNT and Win2K era, when MBR limitations made large volumes tricky to create. Such needed some trickery to address all of that, and in the era the trickery had a cost. This magnified the lots of little files performance penalty. GPT doesn't have this problem. GPT volumes were first in Windows 2003, but their newness meant conservative sysadmins didn't use them much at first.

Old attitudes die hard, sometimes. Especially if there is trauma reinforcing them.

3

We have some much bigger than that (using W2K8) and we've not seen any noticable performance drop offs.

3

You're more likely to hit hardware performance issues long before software related ones.

Consider this,

  • Disk: 10x 1tb sata green disks might outperform 5x 2tb disks. One reasons is that you will have more disk queues (DiskQueueLength) to use with the samller disks. However, keep your OS on a seperate faster volume. The RAID card and it's cache also makes for some serious performance considerations.
  • RAM: also affects the number and size of files that can be opened (PagedPoolSize).
  • NIC: will affect how much data can be written to/fom the server. Consider having your backups run off a seperate NIC. And use adapter teaming to combine the two primary NICs to act as one.

Currently I have two servers. The volume is 16TB and 6TB of data 5million files/directories. At the end of the day, 200GB of data has been changed or added. There are 100 users connecting to the server. And of course these servers replicate the data real time using DoubleTake. No one complains about the server performance.

That said, it also depends on what you want to implement. MS DFS-R on Win2008R2 for example does have limits such as 8million files and 10TB per server/volume as an offical answer. But I've seen reports of 20+TB.

You can also have a read through this if you want... http://www.microsoft.com/whdc/system/sysperf/perf_tun_srv.mspx

5
  • I should also include some more links. You will need to read through them and decide foryourself if any of the hacks are usable. I know that things like disabling last access time are not possible. But defrag once a week is a must. And speaking of which, if you change the block size to 64K, you will have to buy a special defrag program, Windows can only defrag the default 4k blocks. waynes-world-it.blogspot.com/2008/03/… oreilly.com/pub/a/windows/2005/02/08/NTFS_Hacks.html
    – cwheeler33
    Jan 13, 2011 at 20:06
  • You seem to have covered a fair number of points but not the one under discussion. Perhaps you should read the question again. Jan 13, 2011 at 21:06
  • Why does everyone insist on using WD Green drives in storage servers? I've had such bad experiences with WD Green's in 24x7 servers and switched to WD Black's and all my problems went away Jan 13, 2011 at 21:16
  • @John, like I said it's a non issue. Most of the problems revolve around the hardware and a few legacy registry entries
    – cwheeler33
    Jan 14, 2011 at 20:37
  • @Mark, I was trying to illustrate that more disks is better. Some older SATA drives run slower then the current Green drives. And if you compare a WD Black to an RE straight up, you would have to go with the RE. But the cost of 6 RE vs 8 black might still be higher. But the performance on the 8 Black would be higher. Most people already have knowledge that faster drives will work better, but they may overlook the fact that more drives will also speed up performance.
    – cwheeler33
    Jan 14, 2011 at 20:55

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .