Select the right disk for your RAID array

Select the right disk for your RAID array based on how frequently you will use the disk and how close the disk will be to your ear. If the RAID array is in your office, low noise is the number one factor. When you can place the RAID array in another room, you can choose based on a wide assortment of other factors.

Brand

Choose any one of the major brands. The critical thing is finding a replacement disk when one of the array disk breaks. Given the low cost of disks, creating a RAID 6 array with two spares might be easier than trying to immediately find a replacement when one disk breaks. See RAID 6.

Samsung is my favourite brand today for quiet disks at a reasonable price. Seagate is my second choice. Hitachi, Seagate, and Western Digital will be around for a long time supplying replacements. Samsung are selling their disk business to Seagate. The merger of the Samsung and Seagate disks should be excellent but finding a replacement for a current Samsung disk might be harder.

Range

The major brands have several ranges. They have a low power range so you can call your system green even though the lower power usage of the disks may be trivial compared to the total power consumption.

Some brands have a consumer range and an enterprise range. The enterprise disks are expensive and are useful only if you have the disk running hard 24 hours per day every day per year for 5 years. The cheaper disks should work well for typical office use 8 to 12 hours a day, 5 or 6 days a week, for 3 years.

If you build a backup server, the server might be on only a few minutes per day. Most good brands of disk can start 50000 times or once per day for 50000 days or 136 years. Compare that to a NAS device storing incremental backups. The device might power up the disks for a few seconds each couple of minutes or 720 times per day or 262800 times per year. Oops, the disk will break after only 50000 starts or 69 days.

For disks that are on all day in very low power backup devices, low power is important and speed is not important. Your workstations and servers need speed plus the disks are minor components of the power usage. Computers sitting in the same room should be quiet.

My preference is performance first, quiet second, reliability third, and power consumption is not an issue. Choosing the right processor can save you more power than replacing ten disks with their lower power versions. Updating to a modern power supply can save you over a hundred watts, the equivalent of upgrading 50 disks to their low power equivalent.

Model

Usage Samsung disks as an example. At one stage the shops were selling the IJ series along side the LJ series. The LJ series uses three platters to store 500 GB and requires 12 watts of power to drive the whole mechanism. The IJ series uses only two platters to store 500 GB and requires only 7 watts to drive the mechanism. This 5 watt difference is large compared to most model changes. A 2 watt saving is more common.

Using a more efficient processor in a server can save from 30 watts up to 100 watts. Replacing two disks of 500 GB with one disk of 1 TeraByte can save you more watts. Generally the best approach is to use the largest disk that is not at premium price. Often the best choice is one disk down from the top of the range.

An important consideration by model is the expected lifetime of the model. If you need a replacement disk 2 or 4 years down the track, what are the chances of your disk still existing? The newest models can be expected to be on sale longer than models that were released a year or more in the past.

Size

When you are using lots of disks in a RAID array, you can make choices along the lines of 5 disks of 2 TB or 7 disks of 1.5 TB. If the 2 TB disk is selling for a premium price and the 1.5 TB disk is on sale, the 1.5 TB disk might save you enough money for a faster processor or more memory. On the other hand, the 1.5 TB disk option might require an extra SATA controller or push you over the limit towards a bigger power supply.

Exact capacity is another consideration. Some disks vary by a small percent age from model to model and from batch to batch. You can have two models with the same stated capacity, 500 GB in my example case, but vary by a GB or more. The size can vary by hundreds of megabytes from batch to batch. Think about an array where every disk is 500.2 GB. You need a replacement and the new model has only 500.1 GB, a size that will not fit in the existing array. You might get a disk with the same model number number but have 500.175 GB because it is from a different batch. Again the disk will not fit.

Some people create RAID array components with a deliberately small partition to remove the size incompatibility. You might format all your 500 GB disks with a 498 GB partition to allow for slight variation in size of future disks.

Size has another implication. When disks exceed 2.2 GB, they have to change their addressing structure or their sector size. This was know 20 years ago and should have been planned for. I used to talk about it all over the information technology industry but everyone said yeah, yeah, we will do something next year. Now it is many next years later and nobody has done anything. Instead of an advance in disk format, we now have a really backward step that is so shoddy, stupid, and ugly, that the marketing people tried to hide the disaster under the name of Advanced Format.

What happens in AF disks is complicated. The sector size is increased from 512 bytes to 4096 bytes to let disks expand from 2 TB to 16 TB. To maintain compatibility, the disks pretend to use the old sector size. that means a disk cannot write to a sector. Instead the disk has to read the existing sector then change the small bit in the big new sector that represents the pretend sector size then write the big new sector back to disk. The result is the same sort of disaster created by RAID 5 when you use the wrong stripe size.

Seagate, Microsoft, and everyone else involved in the Advanced Format disaster should be treated to something too vile to mention here. The result of their plan is typical of a plan made by a committee. You can imagine the discussion. What is the worst possible outcome? OK, how can we make that happen? Ooh, that sounds ugly. Lets call it Advanced Format so all the suckers out in consumer land think we are doing something good for them.

Speed

Speed is the most important consideration for desktop computers, workstations, and servers. Your time is more valuable than the minor cost difference between a slow and a fast disk.

If the only thing you do is read email or read web pages, a slow Dell or similar cheap computer might be acceptable. Almost everything else deserves a medium speed computer, perhaps a high end HP or an expensive Apple. When you need computers in the workstation range, for all day continual power use, you are usually out of the consumer range. You need a custom built computer or might use a server from a commercial range.

Back when I worked on project X (they will remain anonymous because their mistakes were due to management), the computer purchaser spent from $12000 up to $30000 on (expensive brand) servers and only $1500 for our workstations. At the time $1500 was enough to build a computer faster than their $12000 servers and close to their $30000 servers although with slightly less disk storage. Their $1500 computers purchased from a major retailer were slower than I could build for $600. Some of the computers might have features useful to somebody somewhere but not to us.

Disks are many times slower than modern processors and memory. Spend a big percent of your budget on disks but do not starve other parts of your computer. Most computers are delivered with processors that are too fast compared to memory and disk. You might use the extra processor speed for playing games or editing video. You do not need it for anything else.

RAID 6

You read about RAID 5 all the time. RAID 5 has one extra disk to carry the workload when a disk breaks. RAID 6 adds more spare disks. Take an example using four disks in RAID 5. You get the equivalent of three disks carrying data and one disk carrying parity data used to rebuild a disk if one disk fails. The data and parity are spread over all the disks. Any single disk can break without the array losing data. If s second disk breaks, you lose everything. RAID 6 would add a fifth disk. Any two disks can break without you losing data. You could add a sixth disk and any three disks could break without data loss. RAID 6 can have any number of spares from one up to the limit of your hardware or budget.

Another advantage of RAID 6 is the option to buy lots of disks from the same batch so they are identical. You do not have to find an identical disk when a disk breaks.

Some RAID 5 hardware and software offers the option of inactive spares so you can use RAID 5 in a similar way to RAID 6. The RAID 5 equivalent leaves the spare disks switched off or idle so they do not wear out at the same rate as the active disks. When an active disk breaks, the RAID system activates a spare disk and immediately begins rebuilding the array. This alternative works only when the array is rebuilt before the next disk fails. A better option would be RAID 6 with two active spares and one or more inactive spares.

Solid State Disk

SSDs, Solid State Disks, cost more, produce less heat, less noise, and are faster for reading data. The latest SSDs, including the new OCZ Vertex III SSDs, are too fast for SATA II and require SATA III. SSDs are sometimes slower for writing short blocks of data. SSDs work best for heavy reading of data. A good use is for delivering media files to a Web site. If all the media files fit on SSD, you can read the files from SSD many times faster than they can be pushed through the network to customers. A better use is for storing a heavily searched database. Experiment with SSDs.

Two low priced fast SSDs in RAID 1 will deliver data at close to the maximum speed of their SATA II cables. The rest of your computer system may have difficulty processing the data that fast. If you use more SSDs, your disks will be too fast for the rest of your system and a waste of money unless you have a specially designed server with the right processors on the right motherboard. Test SSD configurations carefully before spending all your money on them.

SSDs in RAID 5 might be slower than SSDs in RAID 1 because individual data accesses can end up as lots of short reads and writes of stripes. SSDs are very slow at writing short sections of data compared to their maximum read and write speeds. The results vary wildly between brands and models because of the individual controller chips and memory chips. Strip side has a big effect. An SSD might write 64 Kilobyte chunks at double the speed of 4 KB chunks. Changing your RAID 5 stripe from 4 KB to 64 KB might make the SSD twice as fast but then impose a massive slowdown on a database that writes 4 KB pages. If you are not buying the absolute most insanely expensive SSD, put a heavily updated database on RAID 5 on magnetic disks instead of RAID 5 on SSD.

SATA II or SATA III

Some new SSDs can use the speed of SATA III. Some new magnetic disks can use SATA II to transfer small amounts of data from the cache in the disk controller to the computer but reads from the magnetic disk to the electronic cache are still far slower than SATA II. There are hybrid disks with SSD included in a magnetic disk but they are expensive and suffer the slow write problem of SSDs plus the slow handling of small data chunks. Many new motherboards have two SATA III ports then four SATA II ports. The configuration is ideal for two SSDs in RAID 1 then several magnetic disks in RAID 5. You then configure your applications to put the small read intensive files on the SSDs.

Configuring Linux and Windows for SSDs is painful because all the files are in directories designed for systems where all the disks are the same speed. There is no concept of fast and slow disks or read versus write. IBM mainframe operating systems had two levels of storage a long time ago and can handle the separation required for the best use of SSD. The simplest split is to buy an SSD that is bigger than you need then dump everything on the SSD except your databases and media files. Based on monitoring of your system you might then move some files from the SSD to the magnetic disk because the files are tiny and continually updated. Your email directory could be a good candidate. Firefox is reported as continually updating hundreds of small parameter files and is another candidate.

My computers

My attempts to put everything in one big RAID 5 array resulted in unacceptable creation times and either unacceptable disk replacement times or an unacceptable slow down while the RAID 5 array was rebuilt. If I tried one big RAID 5 array again, I would make it RAID 6 so I could delay the repair to a more convenient time. With modern large disks, RAID 1 makes more sense. Most of my current computers feature two small SSDs in RAID 1 for the system and two large magnetic disks in RAID 1 for everything else.

For the next size up in storage, I use two SSDs in RAID 1 plus four magnetic disks as two lots of RAID 1 or one RAID 10 array. RAID 6 would kick in when I have a system with many more disks.

My working computers are in the same room and have disks selected for high speed with low noise. Most of the disks are selected Samsung models with Seagate the next choice. Some low power green models are quiet but slow, too slow for regular use.

My backup servers are in another room and use disks selected for reliability plus low power. The low power green disks were sold at a premium price and are now down at prices matching their usually slow performance, making the green disks great choices for use in very low power NAS devices, like the Qnap TS-210, but not for use in high activity servers.