P3R wrote:The scope of the test is clearly stated in the first post and the title itself isn't making any claims at all about what's tested, the results or is in any other way misleading. In my opinion you're grasping at straws when requesting a change of title/subject.
What "straws"? I don't have an agenda here - I just want to ensure that there is no misleading information.
Please address me directly when your complaining about my posts. You did so initially here but I tried to focus the discussion to the other thread (that now on your request seems to have been moved
here) where you also attacked me.
From the beginning, my complaints have been - for the most part - about the conclusions drawn. When I called something misleading, I'm not --specifically calling out
your posts. The comment about "myth" though did come from you.
Perhaps my original tone was a little aggressive. I'll grant you that, and for that you have my apologies. However your comment "that's the internet myth that so many are blinded by" does not really imply a specific focus on QNAP devices and/or limited-bay devices. By citing "the internet" you're literally making the claim as global as possible.
Indeed, in that and some other posts I have been oversimplifying the case, in exactly the same way RAID 10 fans had been oversimplifying the case for several years before this test showed up. For the first time there was testing done showing that RAID 5 and RAID 6 didn't s*ck as bad as these guys have stated over and over referring to the "write penalty". I knew my real life experience wasn't as bad as the RAID 10 bunch claimed it should be but I didn't have the hardware (or to be honest patience) myself to debunk the myth so forgive me for being excited when somebody finally presented this test.
Yes, exactly, and that's precisely why I think it's important that we disclose/state the --whole-- truth, and not introduce --more-- incomplete information.
You called the extensive testing opening this thread "anecdotal reports". Again, English isn't my first language so I looked the word "anecdotal" up to see if the meaning really was as bad as I thought. It was.
There is nothing "bad" with the word anecdotal. Anecdotal means, you heard something somewhere, and you relay it somewhere else in indefinite terms. If you were to go to another storage forum and claim "I know a guy who did tests and found that R5 was faster", that would be a very definition of an anecdote. Links & concrete data avoid that, but I imagine that many people read this thread, then go away thinking, "cool R5 is faster than R10", and any time they then report that info, that's anecdotal.
Feel free to think he didn't do the test you wanted to see and you may think my comments here where exaggerated (which they were) but that was a very low comment to make of the work the OP presented here!
He can do any test he likes. My objection, as it has been from the beginning, is the usage of "it's faster" without context or mitigating considerations.
Individual disk size, availability expectations, network infrastructure, number of concurrent users, monetary budget, type of usage, current and future storage demands are in my opinion all important factors to consider before giving RAID advice so I welcome everyone asking for things like that before making recommendations here. Please accept my challenge to become a more active contributor and try to improve the forum in the future instead of simply fighting my soon 3-year old post here.
I fully agree. And in the context of this forum, where RAID structures are primarily count-limited, the data is a valid, interesting analysis of write performance on consumer NAS devices. What it is --not-- is a general evaluation of RAID-level performance, and while I'm not claiming that was even the intended message, the resulting commentary
was ambiguous and my only goal was to attempt to remove some of that ambiguity.
You know what, you guys asked for facts/data. Well, how about we take a closer look at the OP's own data...? Let's see what happens if we match
capacity instead of spindle count.
To clarify:
4-disk RAID 10 = same capacity as 3-disk RAID 5; 4-disk in RAID 6
6-disk RAID 10 = same capacity as 4-disk RAID 5; 5-disk in RAID 6
8-disk RAID 10 = same capacity as 5-disk RAID 5; 6-disk in RAID 6
SO, let's see how that lines up. From the OP's own data:
qpio wrote:
1. NAS with 8 bays with Atom D2700 dual core @ 2.13 GHz (not a QNAP model, but that 'other' brand, wink wink, psst. the 1813+)
2. QNAP TS-870 with Celeron G550 dual core CPU @ 2.6 GHz
3. QNAP TS-870 with i7 3770t quad core CPU @ 2.5 GHz
4. Disks used (always using disk 1-n when testing with 'n' disks):
disk 1-4: Hitachi Deskstar 1TB 7200 rpm
disk 5-7: WD GREEN 1TB
disk 8: WD GREEN 2TB
5. Disable syncing (echo idle > /sys/block/md1/md/sync_action) -- avoid 10% performance hit
6. Disable thin provisioning -- avoid 15% performance hit
1. NAS with 8 bays with Atom D2700 dual core @ 2.13 GHz
3 disks in raid-5 72.11 MB/s 76.35 MB/s
4 disks in raid-6 113.6 MB/s 110.8 MB/s
4 disks in raid-10 155.2 MB/s 147.7 MB/s
4 disks in raid-5 167.1 MB/s 176.2 MB/s
5 disks in raid-6 154.8 MB/s 135.3 MB/s
6 disks in raid-10 227.4 MB/s 218.8 MB/s
5 disks in raid-5 210.5 MB/s 176.2 MB/s
6 disks in raid-6 175.8 MB/s 165.3 MB/s
8 disks in raid-10 238.9 MB/s 249.5 MB/s
2. QNAP TS-870 with Celeron G550 dual core CPU @ 2.6 GHz
3 disks in raid-5 135.3 MB/s 134.8 MB/s
4 disks in raid-6 133.9 MB/s 133.3 MB/s
4 disks in raid-10 147.2 MB/s 150.4 MB/s
4 disks in raid-5 202.1 MB/s 203.4 MB/s
5 disks in raid-6 207.8 MB/s 211.7 MB/s
6 disks in raid-10 229.3 MB/s 234.8 MB/s
5 disks in raid-5 275.4 MB/s 268.0 MB/s
6 disks in raid-6 270.1 MB/s 271.4 MB/s
8 disks in raid-10 295.2 MB/s 298.6 MB/s
3. QNAP TS-870 with i7 3770t quad core CPU @ 2.5 GHz
3 disks in raid-5 155.6 MB/s 156.1 MB/s
4 disks in raid-6 155.2 MB/s 159.5 MB/s crypted: 134.5 MB/s 127.9 MB/s
4 disks in raid-10 166.0 MB/s 159.4 MB/s
4 disks in raid-5 241.8 MB/s 230.9 MB/s
5 disks in raid-6 245.9 MB/s 221.9 MB/s crypted: 184.1 MB/s 189.1 MB/s
6 disks in raid-10 254.1 MB/s 257.4 MB/s
5 disks in raid-5 325.2 MB/s 316.5 MB/s
6 disks in raid-6 324.2 MB/s 315.8 MB/s
8 disks in raid-10 347.9] MB/s 316.2 MB/s
Summary: A constant and complete victory for RAID-10. Now, does that mean that RAID-10 is "simply faster"? No, of course not. In fact the OP even calls out a number of factors:
(1) when comparing RAID-5 and RAID-6:
RAID-5 and RAID-6 are comparable when you look at the same number of data disks (e.g. 2+1 in RAID-5 compares to 2+2 in RAID-6)
Yes, exactly - so, why not use the same rule with R10 in the first place? Not doing so is misleading.
(3) comparing CPU's:
(b) the G550 and i7 are much faster, both run out of steam beyond roughly 6 data disks in RAID-5 and RAID-6
(4) when considering RAID-10
The G550 and i7 CPU's are more than capable to perform the XOR operations at high speeds.
Indeed, the modern CPUs do a great job of neutralizing
a lot of the dreaded "write penalty".
(4) when considering RAID-10 In the numbers you can see that RAID-10 only gives a slight advantage when running on the Atom CPU. Also, note that only write speeds where compared. The striping advantage of RAID-10 is only effective when reading.
This is misleading. Well actually, not misleading, it's just incorrect. The striping advantage comes into play for writing. It's the mirroring and the reading from both mirrors simultaneously that makes the [primary] difference for read results. XOR 'penalty' is of course also relevant, but less so with more powerful processors, as we've seen.
So it should be very clear to see that the "RAID-10 performance myth" is anything -but- a myth. You just have to have your context right.
Now, absolutely, as it pertains to limited-bay QNAP devices, this consideration may well be less valid. In the context of QNAP users, and this forum and their likely usages, the data presented here is relevant and of interest. But as has been made abundantly clear, there are many factors that influence the overall performance characteristics of the array, and so really the only thing that should --not-- be claimed here is that one RAID level is --necessarily-- ever faster than another, without giving the appropriate context. Which is exactly what I said from the start.