2

To make a long story short, I am wondering if it is unwise to use large thin .vmdk's for a nas VM? I am planning on backing up to external 2TB drives (very small business), keeping one offsite at all times.

I could use a poweredge 840, but the fileserver is so lightly used, it seems like a waste. The company has an ESXi server that could store the fileserver disks.

Regarding the backups, would I be better off using cron+rsync and passthrough to USB using the native disk, or should I make a datastore and backup the .vmdk files? Also, there isn't any budget for software.

SkinnyGeek1010
  • 446
  • 5
  • 16

2 Answers2

2

You can create a VMDK as large as you want -- I don't believe VMWare imposes a limit. It's as safe as any other VMDK image, with perhaps a slight increase in risk as it's covering more physical area on the disk (greater chance for a sector to go bad. That's why your VMWare box has a RAID controller, right?).

The two caveats I would point out are:

  1. If you use a sparse (growing) image instead of a pre-allocated one your VMDK may not be contiguous. This can create performance problems on two ends: Writing data will take longer (because the VMWare server needs to allocate physical disk space), and retrieving data will take longer (because the head has to seek over the "gaps" in the disk).

  2. If your file server is heavily used you may create disk contention: People trying to read/write files will be competing with all the other VMs trying to do whatever they do. Depending on disk speed and overall VM activity this may make a noticeable dent in performance for the file server or other virtual machines.

Sometimes buying the El-Cheapo standalone file server (or a NAS device, or putting a copy of FreeNAS on an old box and giving it big disks) is a better solution - Consider your workload before virtualizing.

voretaq7
  • 79,879
  • 17
  • 130
  • 214
  • 6
    I wouldn't be surprised if they start charging per kilobyte of VMDK space used, just to screw the users more on licensing fees. – Tom O'Connor Jul 26 '11 at 15:16
  • Great points. I was under the impression that 4.1 fixed speed issues with thin disks, apparently I need to research this further. I am definitely leaning towards the dedicated box now, and I could get a raid card cheaper for that. – SkinnyGeek1010 Jul 27 '11 at 01:02
  • Also forgot to add.. They write an average of 300-400mb/day and read around 100mb or less. Backups at night would be 1gb or less. The archives (print jobs, tiffs,PDFs) reach around 30gb/mo and they have around 450gb of archives now. – SkinnyGeek1010 Jul 27 '11 at 01:20
  • 4.1 made major improvements in disk speed, but it doesn't overcome the physical limitations of rotating disks and moving heads -- sufficient disk IO on any VM hypervisor will impact the overall performance of guests running on that host. – voretaq7 Jul 27 '11 at 03:28
  • 1
    Regarding disk size - make sure when you format your data store you keep in mind the size of sectors. Otherwise you'll run into per-file VMDisk limits. – Bart Silverstrim Jul 27 '11 at 15:52
  • 1
    By the way, there is absolutely a 2 TB per-VMDK limit even in the most recent versions of ESX(i). – Joe Jul 31 '11 at 00:41
1

In my opinion both the alternatives you present for backup are a bad idea. I highly doubt hot-plugging USB hard drives was in the design brief when VMware designed the datastore system. Yes you can dismount datastores, but I'm not sure ESXi supports using USB devices as datastores anyway (I think you'll find the embedded OS can't mount volumes from USB devices).

USB passthrough for storage devices is available in ESXi, but you have to shut down the VM to detach it cleanly. Again, using this mechanism for routine backup is a bad idea as it's highly inconvenient to do cleanly. I've also experienced data corruption while attaching a USB HDD to a CentOS 5.5 VM on ESXi 4.1 for a migration... I wouldn't trust it for backup.

A cheap NAS gets my vote.

Regarding using large VMDKs I'm not aware of any specific risks of that size, but it doesn't seem like a great idea to use a thin 1TB vmdk for a NAS, especially since there will presumably be a lot of writes.

Alex Forbes
  • 2,452
  • 2
  • 20
  • 26
  • Wow, I didn,t realize USB disks had so many problems. I was going that way to passthrough the disks directly to the vm, but that's too much trouble. I didn't specify but the datastore would be a sata hot swap tray, not USB. Would this have the same issues hot swapping to the 2nd drive? – SkinnyGeek1010 Jul 27 '11 at 01:07
  • Even with SATA hot swapping cleanly requires OS support. I think this is more likely to work than USB, but having never tried to hot-swap whole volumes on an ESXi host I'm not sure how clean it would be. If you have a hardware RAID controller and the volume was a RAID1 mirror (so that the OS could see the physical volume the whole time), I think it would work. You'd just want to ensure that nothing was writing to the volume at the time of removal, and that it resynchronised properly when reinserted. – Alex Forbes Jul 27 '11 at 08:45