3

While chatting with a sysadmin colleague of mine about setting up a small server for me, I found out something interesting - most servers these days apparently contain an internal USB port which can be used for attaching a boot flash drive. On one hand this makes sense - the OS doesn't need to use its disk much, and all the data is being kept on actual fast hotswap drives. We've successfully eliminated the need for an extra drive bay.

On the other hand - flash drives? Really? Is this safe? Seems so weird. Is this really common practice today?

duenni
  • 2,939
  • 1
  • 22
  • 38
Vilx-
  • 791
  • 4
  • 13
  • 25

3 Answers3

2

It is pretty common practice today, especially for virtualization hosts (ESXi, Hyper-V, etc.).

joeqwerty
  • 108,377
  • 6
  • 80
  • 171
  • What about non-virtualized scenarios? Would it be OK to do it then? My idea is OS and all software on the flash drive, but data on separate disks. – Vilx- Jun 27 '19 at 12:14
1

Standard server often come with OEM OS on flash card. It's usually on MicroSD card.

Often you can mirror it too, as you can have two SD card on the controller.

On the other side the most practical use is for having a hypervisor on it, like ESX. The hypervisor, even if it crash can leave other server in the pool handle the VM witgh a proper HA configuration, and re-installing a ESX inside a SD card take 30m in the worst case.

I would be careful to do a flash setup if it's a standalone scenario.

yagmoth555
  • 16,300
  • 4
  • 26
  • 48
  • Hmm, my use case would actually be a standalone scenario, no virtualization (my needs are humble). What would you say about installing the OS on a flash drive then? The other disks would contain only pure data, no software. – Vilx- Jun 27 '19 at 12:12
  • @Vilx- You can do whatever you want honestly, but just keep in mind if it crash, how much time it got to recover. SD card tend to crash way more often than standard disk. – yagmoth555 Jun 27 '19 at 13:46
  • Kinda, but on the other hand - they're dirt cheap and easy to clone. I could have full system backups ready to swap at any time. – Vilx- Jun 27 '19 at 17:41
  • @Vilx- Yes, it's why it's a choice of yours, like I told some server even offer dual card setting to have a RAID-1 of them. It all depend of what is hosted too, as like ADDS, the directory is local on the disk usually, which can be unsafe to store, but for like a webserver it's not a problem to bring back one – yagmoth555 Jun 27 '19 at 18:25
1

We have thousands of servers running VMware's ESXi off SD cards, I believe you can also run Photon Linux and possibly Windows Nano off SD/USB cards too.

Chopper3
  • 100,240
  • 9
  • 106
  • 238
  • What about non-virtualized scenarios? Would it be OK to do it then? My idea is OS and all software on the flash drive, but data on separate disks. – Vilx- Jun 27 '19 at 12:13
  • Well the Windows Nano isn't virtualised - there's nothing to stop any OS running on SSD/USB, the issue is around write-wear, you don't want your OS writing too much, or your apps for that matter, if you can control your writing then you could run anything that way. Does that make sense? – Chopper3 Jun 27 '19 at 13:59
  • Well, yeah, but that then moves the question to "how much is too much"? And I guess there's not going to be a good answer to that and I'll just have to use my best guess. But what about stability? Will the USB hardware/software manage a 24/7 attached drive without hickups? It's mostly used for removable storage after all. – Vilx- Jun 27 '19 at 17:40
  • We've ran with SD card booted ESXi now for something like 8-10k server-years and had 2 card failures, which I think is pretty good. But obviously if you run a non-SD/USB-aware Linux distro or 'Normal' Windows then this number will come crashing down. That said there are some reasonably high-wear USB drives about, plus if your server had an m.2 slot you could use an Intel Optane based drive and they can be written to far more than regular SSDs or flash. – Chopper3 Jun 28 '19 at 11:14