Datacenter ideals

Should we use blades, racks, or towers?
Depends on the business.
Horizontally oriented computers are generally less deep in relation to width than towers are wide in relation to height.
Towers are cheapest and are good if you have a single server or home computer. I think they have good IO and expansion capabilities. They take more space though, and mean a lot of cabling. I’ll edit this part of the post later after I read about towers, racks, and blades.
Racks are 19″ wide and are 1U(1.75″), 2U(3.5″), or 4U(7″) in height. They are mounted on a rack, which is often 42U (approximately 73″) tall. They mean less cabling than towers, take less space, and are considered by some to be a compromise between the advantages of blade servers and the advantages of rack servers.
Blades are stripped of certain features such as IO, power supplies, and other things to save space. The minimum height of a rack is 1U – 1.75″. Blades, being stripped of some features, can be thinner. Racks and towers can run on their own, while blades need a blade enclosure and cannot be used without it. The blade enclosure provides power supplies and IO which are shared by the blades. Since IO and power supplies and other things are shared, the system has better density than towers or racks. Blades are more expensive than racks or towers, though. Blades are often vertically oriented, but they could be horizontally oriented.
So, racks are best for general business. Blades are best for the enterprise, datacenters, HPC, and cloud computing.

Converged infrastructure

So what? There are normally much fewer storage nodes than compute nodes, so in a converged infrastructure, to combine the best of both blades and racks, the storage nodes should be either one 4U rack or two 2U racks, while compute nodes should be blades in a single blade system.

Storage

RAID stands for Redundant Array of Independent Disks. It is the most common type of storage array. RAID0 is the most basic RAID. It uses striping. RAID1 uses mirroring, which means having data on one or some disks and backup data on other disk(s). Three RAID designs use single dedicated parity: RAID2 uses bit-level striping, RAID3 uses byte-level striping, and RAID4 uses block-level striping. RAID-DP uses double dedicated parity. RAID5 uses single distributed parity, and RAID6 uses double distributed parity. Other RAID models include RAID5E, RAID5EE, and RAID6E, in which the E stands for hot-spare drive. There are also nested RAID levels, my favorite of which is RAID10, which is a RAID0 of RAID1’s. The other, which is less common, is RAID01, which is a RAID1 or RAID0’s. There are also RAID50 and RAID60, and you should be able to guess what they are. All disks in a RAID must be the same size. Otherwise, the additional space of the larger disks will be wasted. JBOD (Just a Bunch Of Disks), on the other hand, can be heterogeneous. JBOD doesn’t have redundancy though, so there might be a single point of failure, at least with SPAN/BIG which uses logical volume manager. There is also a nested design cakled JBOD RAID N+N, which is a JBOD of RAID’s. There is also Synology Hybrid RAID, which combines the best of both JBOD and RAID. It has redundancy and striping like a RAID, and similar to a JBOD, you can add disks that are larger than the current disks. You can’t add smaller diska though; There is a minimum size which is the size of the original/basic disks. Another disk array type is the MAID(Massive Array of Idle Drives). It turns off disks that are not in use. That saves power, and also increases durability, and other things. A MAID is cheaper than a RAID. It is slower than a RAID or JBOD, because in a MAID, when disks are needed, they have to be turned on; in a RAID or JBOD, the disks that are needed are already on, so that saves time but wastes power. A MAID also has less fault-tolerance than a RAID. A MAID is used for nearline storage that is often “write once, read occasionally.”
There are HDD’s and SSD’s. HDD’s are much cheaper and have a great capacity, but they perform poorly. HDD’s use much more electric power. The write speed is good though. SSD’s are much faster, but they are expensive and don’t hold as much. SSD’s also wear out upon writing and don’t have much better write time than HDD’s. SSD’s have much better read time, though. Hybrid drives can be either dual-drive or solid state hybrid drives(SSHD’s). SSHD’s are more tightly coupled and are a single volume with a single interface, best would be Serial ATA(SATA). Dual-drives have separate interfaces, best would be Peripheral Component (PCI) express for SSD and SATA for HDD.
So, hybrid HDD and SSD for an array shouldn’t be a RAID. It would be a hetergeneous JBOD. Or maybe a JBOD of two RAID’s: one of HDD’s and one of SSD’s. Or even better, a JBOD with a MAID of HDD’s and a RAID of SSD’s. Since HDD uses more power and an SSD is faster, having the MAID be of HDD’s and the RAID be of SSD’s would be better as it would save more power and mean higher peformance than having the other way around(MAID be SSD’s and RAID being HDD’s). Use either all-flash or better hybrid array for datacenter; Not HDD for that.
You could have the SSD and HDD be a single volume, but that’s not a best practice, because the computer wouldn’t know what files and blocks are in SSD and which are in HDD. There would be sharp contrast in performance between different areas in the volume due to some areas being in HDD and some being in SSD.

Conclusion: anything I missed? Say so in the comment. I might make some edits and add some stuff later.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: