I'm currently running 2x 3TB Red drives and a 3TB toshiba DT01ACA (7200rpm) parity. I did ok on Black Friday, grabbing a 5tb HGST NAS drive from frys ($100, score) and a Toshiba 5TB X300. Both 5tb drives are preclearing now; the HGST is running around 155 MBps and the X300 is at about 170.

 

I'm torn as to which drive I should set as my new parity drive; I go back and forth on having the faster drive holding data for faster access and having faster parity performance (assuming that the speed of parity drive is the bottleneck in the array). I have an SSD cache, so it's mover speed that I'm considering, I guess.

 

Any thoughts? Am I thinking too hard about this?

Writes to the array can't go faster than the slowest component.  So if you are writing to a 5400rpm Red drive it won't matter if your parity drive is 7200rpm.  Still, the best practice is to have your parity drive be as fast as, or faster than anything else in the array.  With your current set of drives it really won't matter much - but you could still add 7200rpm data drives later.  In the meantime your parity checks might be a little faster with the 7200rpm as parity, or if you are writing to multiple drives simultaneously?

Use the faster drive for parity.

Parity is the bottleneck in writes because it does a read-write-read, iirc.

Ideally it should be twice as fast as the data drives.

 

The parity drive should have fast access times which is highly dependent on spindle revs.

The highest throughput is also beneficial, as it will finish read/write earlier, also dependent on spindle revs.

I noticed a big buffer/cache is also positive in terms of throughput.

 

I'm collecting some drive info in the wiki.

Would you mind contributing to the table?

 

Use the faster drive for parity.

Parity is the bottleneck in writes because it does a read-write-read, iirc.

Ideally it should be twice as fast as the data drives.

That is probably a little overstated.

 

The way writes to the parity array actually work, for both normal and turbo mode, is explained in the Turbo write thread.

Writes to the array can't go faster than the slowest component.  So if you are writing to a 5400rpm Red drive it won't matter if your parity drive is 7200rpm.  Still, the best practice is to have your parity drive be as fast as, or faster than anything else in the array.  With your current set of drives it really won't matter much - but you could still add 7200rpm data drives later.  In the meantime your parity checks might be a little faster with the 7200rpm as parity, or if you are writing to multiple drives simultaneously?

 

Thanks!

 

Use the faster drive for parity.

Parity is the bottleneck in writes because it does a read-write-read, iirc.

Ideally it should be twice as fast as the data drives.

 

The parity drive should have fast access times which is highly dependent on spindle revs.

The highest throughput is also beneficial, as it will finish read/write earlier, also dependent on spindle revs.

I noticed a big buffer/cache is also positive in terms of throughput.

 

I'm collecting some drive info in the wiki.

Would you mind contributing to the table?

 

Thanks. I'll update the Wiki when the preclear cycle finishes.

... I go back and forth on having the faster drive holding data for faster access and having faster parity performance ...

 

You're definitely over-thinking this.  First, although there is a slight difference in the access time with a 7200 vs 5400 rpm drive; it's a VERY minor difference (a couple msec).  The actual transfer speed from your drives -- which is what I presume you're thinking about -- will be effectively IDENTICAL, as both drives can stream data faster than a Gb network can -- so unless you have a 10Gb network your transfer speeds will be limited by your network; not by the drives.

 

Where that faster few msec of access time comes in handy is on a parity drive IF you're doing multiple writes at the same time => e.g. from two different clients.  The comment "... if you are writing to a 5400rpm Red drive it won't matter if your parity drive is 7200rpm ..." is true IF that's the only write; but if you're writing to two DIFFERENT drives (regardless of their speed), then the access time of the parity drive absolutely makes a difference in performance.

 

HOWEVER ... with cached writes, even that doesn't matter, since the writes are to the cache and parity isn't being updated during those writes.  If all of your shares are cached, then it really won't matter from your perspective, since the actual writes to the array will occur when you're not actively using the array.

 

Id be more inclined to put the more reliable drive in the parity spot. While the X300 might be ok, the HGST NAS drives have more examples of being reliable from what I recall.

Id be more inclined to put the more reliable drive in the parity spot.

Genuinely curious about this. What is your reasoning? What are the consequences of parity failing vs. data drives?

It would seem completely irrelevant where you put your "most reliable" drive => it's a fault tolerant array; and when a drive fails you're going to replace it.  No matter which drive fails, the consequences are the same -- NO data loss for one failed drive (or two failed drives if you're using dual parity) -- and the loss of the data on the failed drives if more than that fail.  In fact, in the latter case, you'd WANT one of the failed drives to be parity, since that wouldn't result in any lost data on that drive [just the other failed drive(s)].

 

... and from a performance perspective, you'd want the drive with the best access time to be the parity drive; so if you have multiple simultaneous writes they would be as fast as possible.

 

The best choice, of course, is to use ALL highly-reliable drives 

My recommendation really came with no concrete evidence, just my personal opinion because no matter which disk you write to (disk1, 2, 3, 4) you will be writing to parity as well. So it stands to reason that the parity drive will have more writes to it then any other drive.

 

That being said, reads will never come from the parity drive (except parity checks) and only data disks, so who knows.

My recommendation really came with no concrete evidence, just my personal opinion because no matter which disk you write to (disk1, 2, 3, 4) you will be writing to parity as well. So it stands to reason that the parity drive will have more writes to it then any other drive.

 

That being said, reads will never come from the parity drive (except parity checks) and only data disks, so who knows.

 

There have been many studies of non SSD drive failures, none has found a correlation between drive usage/duty cycle and failure rate.

 

SSDs have shown that writes lead to failures.

 

Anyone trying to find or buy "highly-reliable drives" for array usage does not understand the basic reasoning behind RAID vs SLED. RAID is Redundant Array of Inexpensive Disk. SLED is Single Large Expensive Disk. RAID always wins, and thus drive reliability is NOT improving. Large consumers of disk drives are asking for lower costs, even if that means lower reliability.

My recommendation really came with no concrete evidence, just my personal opinion because no matter which disk you write to (disk1, 2, 3, 4) you will be writing to parity as well. So it stands to reason that the parity drive will have more writes to it then any other drive.

 

That being said, reads will never come from the parity drive (except parity checks) and only data disks, so who knows.

As I said here in this thread.

BackBlaze, Google, and other large users of drives tend to go with the lowest cost drives, since their systems are both very redundant and thoroughly backed up.  That's not a bad strategy -- a lot of UnRAID users use the same one ... always buying the lowest cost/TB drives, and simply replacing them when they fail.

 

I prefer to use higher quality drives -- even if they cost a few extra $$/TB -- to try and minimize the number of times I have to "mess" with drive replacement => but there IS a cost associated with doing that; and whether or not it's really cost effective is certainly debatable.

 

But regardless of what you focus on when buying your drives, it simply doesn't make any difference which drive you use as parity, except that if you have mixed drive speeds it's generally better to use the faster drive as parity.

 

I tend to use the cheapest crap as data drives and a decent drive for parity.

 

At the moment, that's a Seagate 8TB Enterprise NAS for parity, and the data drives are shucked 8TB WD EZZX drives (sort of Reds but not) and a recertified 5TB Red.

...  sort of Reds but not ...

 

??  If they're not Reds, they're not helium-sealed units.  I'd say that's a very significant difference !!