Personal Best Practice Building a SAN?

Question

linux

I especially rarely work with this, but now the task has arisen to organize storage. There is a variety of hardware from the manufacturer Dell, and a certain number of SaS disks, 150-300 gigabytes each.
There are several servers in which you can insert up to 10 disks, with raid controllers allowing you to build raids 1, 0, 01, 10, 5, 6 etc.
I would like to build an iscsi cluster (I don’t know the correct names) to connect to it and store on it virtual machines for migration between servers (meaning virtualization on the server itself that plays the role of storage) - it sounds cool, but I don’t even know if this is applied and how.
How to organize such things in general, how to configure storage fault tolerance, which system is better to use, does it make sense, or any solutions for using virtualization or containerization for this.
For me, this topic is new, so far I have only organized Centos 7, + Raid5 5Disk 250GB + LVM (two groups),
I don’t know what to do next. I think to raise another exactly the same system and use DRBD to synchronize data.
I ask for advice pictures, playful stories.

Answer 1

Rules such
1. RAID5 - last business. Statistics show that the probability of a disk failure during a rebuild is over 70%. The failure of the second disk usually means that the array can be discarded.
2. Creating a RAID using a hardware controller is a bad idea. You thus create a single point of failure (this same controller). If the controller fails, you will have to look for a replacement (the exact same controller with the same firmware), otherwise it will be very difficult to extract data from the array.
3. Centos contains the dorky NetworkManager (and systemd, but the latter can still be survivable). But network management through dbus is hard to survive. Either break NetwokManager, or switch to a normal distribution - SuSE
4. Classic clustering for virtual machines and migration, option 1.
Two servers, RAID10 arrays on the servers. Disks are given to the world via DRBD in active-active mode.
A distributed cluster system OCFS2 runs on top of the disks.
The cluster is managed by the pacemaker.
RAID10 is an array that combines relatively high reliability (at least half of all disks in the array must fail) and relatively high speed (due to splitting the file across disks)
DRBD is such a network RAID1. Data is replicated between nodes during operation, so stopping one node in a pair does not stop storage.
OCFS2 is a cluster file system that allows a bunch of machines to see one block device as their local file system. It has file-to-memory mapping and global locks. And the most useful thing for virtual machines is the ability to create reflink. Reflink is an instantly created copy of a file (virtual disk image). For a new copy, OCFS2 will only store the differences from the original file.
pacemaker is a cluster manager + load balancer.
How to raise it all?
1. Install OpenSuSE 42.1 on all machines
2. Build RAID10 via LVM (there is a GUI for this - Yast is called)
3. In the same YAST, raise DRBD
4. Install the ha-cluster-bootstrap package 5. Make ha-cluster
on one of the nodes -init and follow its instructions
6. Do a ha-cluster-join on the rest of the nodes.
This is in general terms, ask questions.

Answer 2

as a less fancy option: freebsd+zfs+hast.

Answer 3

freenas+zfs+raidz2

Answer 4

SAN is a little more complicated than just a bunch of servers with disks included in the network :-)
> I would like to build an iscsi cluster (I don't know the correct names) to connect to
> it and store virtual machines on it for migration between servers (there is in view
of > virtualization on the server itself, which plays the role of storage)
You mixed a whole bunch of concepts into a bunch, sorry for the pun :-) iSCSI is, if on your fingers, a network version of the SCSI protocol, which is designed to access block devices (storages) via network.
Migration - it has absolutely nothing to do with iSCSI, and is implemented by a whole bunch of technologies, not only with the help of iSCSI storages.
"virtualization on the server itself, which plays the role of storage" is what is fashionably called "converged system". And it is implemented by a separate bundle of technologies and products that do not necessarily include iSCSI.
In general, depending on what exactly you want to get in the end, the scenarios and the software used will be different. Block storage, migration and convergence are things, of course, related, but not the same thing. Something like Karl Marx and Friedrich Engels, or, if you like, Lenin and the party ;-)
I will answer on the last point, on the convergent system. If you want to achieve just that, then you can look at ceph for free and with big hemorrhoids. Linux can be installed on servers, some hypervisor (such as KVM), as well as ceph components can be installed on Linux. And arrange a convergent system out of all this.
There is an option to implement storage on virtual machines. Those. first of all, a hypervisor is installed, a number of virtual machines are deployed on it, into which disks from servers are presented, and then a shared storage is stitched together based on these virtual machines. This is what EMC ScaleIO does, for example. Compared to Ceph, ScaleIO is much simpler and more logical, has components and access drivers for Linux, Windows and drivers directly for ESXi (i.e. no need to fool around with network protocols like iSCSI or NFS). But there are also disadvantages - it is officially free only for test environments. Accordingly, you can use it in production only at your own peril and risk, there will be no support from the manufacturer. And it is not very well documented in terms of its internals. There are manuals "how to deploy" and "how to manage", but there are no manuals "how it can break, why, and how to fix it" - alas, no. Here it’s only either by field experiments, or with the help of the community (it’s small, but it exists), or to score and use something else :-)