New build - can't boot properly and database locked

[Sir.Robin]

The drives are speced to operate at 60 degrees C. Although i agree that 50 is a tad high. Turned up the fan speed today :)

Connection is SATA and ACHI enabled. Connected to the motherboards SB850.

As i wrote a couple of posts up... all drives give the same speed. 12000 kBps.

Maybe i can get some other drives to test with... come to think of it.. i have a Raptor laying around. Gonna test it later today.

 

[paleoN]

Thats around 12MB/s per drive.

Isn't that very low even with those old drives? I'd expect at least double at a minimum. Which would only be half of the 50MB/s.

Now... i deleted my array and created a new one. Single drive. UFS.
And guess what? I get the same write speed (per disk) on this array. 12200kBps.

I'd still test one disk raw and see if there is any difference in reads or writes.

 

[Sir.Robin]

Yes i think it is low... but at the same time the lowest read is about 25M/s on the inner tracks.

How do you test it "raw"?

 

[paleoN]

What I said a few posts [post=29151]back[/post]:

This command will write to the hard drive and will delete data. Only run it on disks not part of an array.

Code:

dd if=/dev/zero of=/dev/ada0 bs=2048k count=50k

dd if=/dev/ada0 of=/dev/null bs=2048k count=50k

That will essential give you your raw performance bound by either the disks themselves or what they are connected through. If you still think you might have a disk that's pulling down the rest of your pool then run the above command for each disk waiting until the previous test finishes.

Then you should run it on all disks all at once. Ideally the performance should be the same as a single disk.

I almost forgot. You can try this as well:

Code:

diskinfo -vt /dev/ada0

 

[Sir.Robin]

sweet command :)

[root@NAS01] ~# diskinfo -vt /dev/ada0
/dev/ada0
512 # sectorsize
500107862016 # mediasize in bytes (466G)
976773168 # mediasize in sectors
0 # stripesize
0 # stripeoffset
969021 # Cylinders according to firmware.
16 # Heads according to firmware.
63 # Sectors according to firmware.
3PM1PGLM # Disk ident.

Seek times:
Full stroke: 250 iter in 5.205713 sec = 20.823 msec
Half stroke: 250 iter in 3.762011 sec = 15.048 msec
Quarter stroke: 500 iter in 6.107419 sec = 12.215 msec
Short forward: 400 iter in 2.004750 sec = 5.012 msec
Short backward: 400 iter in 2.128807 sec = 5.322 msec
Seq outer: 2048 iter in 0.145437 sec = 0.071 msec
Seq inner: 2048 iter in 0.158568 sec = 0.077 msec
Transfer rates:
outside: 102400 kbytes in 1.682233 sec = 60871 kbytes/sec
middle: 102400 kbytes in 2.023884 sec = 50596 kbytes/sec
inside: 102400 kbytes in 3.272459 sec = 31291 kbytes/sec



Ah... now i see the difference in your other commnad: dd if=/dev/zero of=/dev/ada0 bs=2048k count=50k
Defines the disk rather than the volume :) captain obvious...

Anyhow, i get the same transfer rate. approx 12000kBps :|

 

[Sir.Robin]

Testing with newer drives (WD RE3 1TB and Hitachi something 2TB) gives me a raw write of 110MB/s and 120MB/s.

Thats just an TEN fold increase in performance :D

So the old Seagates are just sucky. Worse than i would have beleived if someone told me :-|
Or maybe theres an issue with NCQ/ACHI between the controller and these old drives.

Array test comes later....

 

[Sir.Robin]

RAIDZ1 (5 drives single parity):

[root@NAS01] ~# dd if=/dev/zero of=/mnt/vol01/tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 382.414979 secs (280779228 bytes/sec)
[root@NAS01] ~#
[root@NAS01] ~# dd if=/mnt/vol01/tmp.dat of=/dev/null bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 251.636088 secs (426704227 bytes/sec)


:)

 

[paleoN]

I happen to have a 320GB 7200.10 PATA drive still in a desktop. A 7.2GB file transferred via FTP to the NAS averaged 64MB/s. The same 7.2GB file transferred from the NAS averaged 40MB/s. Is the 7200.10 really that much better than the 7200.9?

107374182400 bytes transferred in 382.414979 secs (280779228 bytes/sec)

Modern drives, modern performance! But then I assume you were planning on replacing the other drives based on the rest of your hardware anyway.

 

[Sir.Robin]

I still think the 7200.9's should provide better than they do for me. The write speed is awful compared to the read speed i get from them.
My best guess is that there is some bug or something between these older drives and the onboard SB850 ACHI capable SATA controller.

Originally i bought new hardware for this NAS but the drives i were planning to use at least the first year or so was the 7200.9's... not anymore though ;)

 

[Sir.Robin]

Again, Thanks a lot for the help and useful commands :)

 

[paleoN]

My best guess is that there is some bug or something between these older drives and the onboard SB850 ACHI capable SATA controller.

Originally i bought new hardware for this NAS but the drives i were planning to use at least the first year or so was the 7200.9's... not anymore though ;)

You could always try them in IDE mode. Of course you would have to take the new drives out first. Have you planned for your future disk upgrades/storage needs? I noticed you started out with raidz2 and dropped down to raidz1. Though that may have been just how many disks you could scrounge up.

Again, Thanks a lot for the help and useful commands :)

No problem. Oh and you want your pool to be 4k formatted if it isn't already.

Code:

zdb | grep ashift

9 means 512b and 12 is for 4k.

 

[Sir.Robin]

I already tried in IDE mode several times. Performance went further down by about 50%.

My current storage needs are under 500GB (but i just bought a video cam)... So my plan were to use these 6 500GB's for now, then later exchange them with 1 or 2TB disks.

I'm still going for raidz2, but as u already guessed... i didnt have enough drives available for now... so a raidz1 with 5 drives were the closest i could get for now.
Still testing... so no hurry :)

Would this be the correct way of doing things for 4k format? :

http://blog.monsted.dk/?q=node/1

 

[paleoN]

If you were doing it from the cli. With FreeNAS you can just check Force 4096 bytes sector size after you pick ZFS for the volume. Depending on the hard drives you have FreeNAS may use 4k anyway.

You just wouldn't want 512b on 4k native drives. It kills write performance. See the 3rd code box of hexland's [post=24532]post[/post] in the benchmark sticky.

 

[Sir.Robin]

Doh! :)

Write 5x drives in raidz1 4096 sector:

[root@NAS01] ~# dd if=/dev/zero of=/mnt/vol01/tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 346.741312 secs (309666540 bytes/sec)
[root@NAS01] ~#