1

I'm kind of a scientific computing (much focus on fast CPUs for distributing big computations like physics simulations and algorithm optimizations, signal and image and video processing and so on) kind of guy who quite new to working with servers and would like to learn more.

Basically if I would have a blade center with 8 or 16 identical blade machines (with just one or none of them having a HDD) would it be possible to remote-boot all the other ones from that one HDD or some other central disk? Or is it maybe so that they share all the disks somehow. Basically what I would want to avoid is having to buy one HDD for each blade to boot locally if they are all in the end supposed to be copies of each other using the same software and so on.

Maybe this is a bit too broad or basic question for this site and what I really should be asking for is some crash course in how to install, boot and configure blade servers?

TomTom
  • 51,649
  • 7
  • 54
  • 136
mathreadler
  • 115
  • 6
  • Why distributed file system? That has nothing to do with distributed file systems and is a standard case of ISCSI and images. – TomTom Jan 17 '16 at 14:39
  • I needed to pick some tags and could not find any better. I think I searched for network-boot, network-install and some other words before. Feel free to edit to anything better if you want to. – mathreadler Jan 17 '16 at 15:30

2 Answers2

1

Using Linux, to boot the blades you need PXE network boot, to boot the operating system.

You can also boot via USB. Most blades I've seen have USB ports on the front or inside the blade.

After booting, if you want to share storage, you'll need to set up a network storage device using NFS or iscsi.

Mike Marseglia
  • 913
  • 8
  • 18
  • So I could install the OS on a scsi drive which I plug into one of the blades and then use PXE network to boot from the same drive? Or do I need to do PXE booting from some external machine? – mathreadler Jan 17 '16 at 15:35
  • 1
    PXE boot happens from another machine. You could install one hard drive into one blade, install the OS, and configure it for DHCP and network boot. The other blades would boot from network, receive an IP address via DHCP and PXE instructions from the first blade. See this tutorial for Debian https://www.debian-administration.org/article/478/Setting_up_a_server_for_PXE_network_booting – Mike Marseglia Jan 17 '16 at 16:36
  • If you are using iscsci, you can probably boot from the iscsci lun – Sobrique Jan 18 '16 at 08:29
  • @Sobrique you would have to set up a unique iSCSI boot lun for each blade, correct? – Mike Marseglia Jan 29 '16 at 15:15
1

In this scenario, with a set of ESXi servers, we have used UCS blades. These attach to shared NFS data stores when operational, but have the same constraint in terms of internal capacity.

So we boot from Fiber Channel over Ethernet presented LUNs from the Netapp that provides the shared storage.

Technically it is separate boot devices, with very slight differences of host persona (hostname, ssl cert, ssh keys etc.) but because of deduplication, the data on disk is really small. (It is pretty similar to your goal)

Overall though, whilst this is quite a flexible and scalable system, it probably doesn't end up being the cheap option.

Of course, depending what you are using your blades for, internal storage might be the best choice. There's a reasonable number of data oriented applications that benefit hugely from local disk type systems. (elasticsearch, hadoop etc.)

Sobrique
  • 3,747
  • 2
  • 15
  • 36