Something Random

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Does it normally take a long time for a software based RAID 5 array to be built? I just setup OpenFiler with their software RAID using my five 1.5TB drives as a single RAID 5 volume. The array is busy building for the first time and it's been running since 6:00AM Sunday morning and it's only at 83% complete (~30 hours). I'm assuming it does take a while to initially build the array for the first time? The activity lights have an interesting pattern with 4 of the 5 drives blinking fast and one staying a bright solid...but it shifts around.
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,719
Location
Horsens, Denmark
Does it normally take a long time for a software based RAID 5 array to be built? I just setup OpenFiler with their software RAID using my five 1.5TB drives as a single RAID 5 volume. The array is busy building for the first time and it's been running since 6:00AM Sunday morning and it's only at 83% complete (~30 hours). I'm assuming it does take a while to initially build the array for the first time? The activity lights have an interesting pattern with 4 of the 5 drives blinking fast and one staying a bright solid...but it shifts around.

Yup, that is the expected behavior. I've had arrays take a week to rebuild. The drive that is pegged is the one storing the CRC data, it will cycle through all the drives eventually.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
I think I'm finally done with this Gigabyte GA-M57SLI-S4 motherboard with an nvidia nForce 570 chipset. I was trying to repurpose my old desktop computer with my Athlon X2 4600 plus to run my NAS figuring I already owned it and it should be more than speedy for a NAS.

I just went through building my RAID 5 array (software based) and I was doing some heavy I/O testing with iSCSI. I was seeing hundreds of buffer I/O errors reported on the console. After viewing the array in the web-based console, several of the drives are no longer found. It was like they disconnected from the SATA bus. I reboot the NAS and my whole RAID 5 array is no longer recognized (even after a cold boot). Basically the 40 hours of initializing it were wasted, but fortunately I had no data on it because I'm just testing. I deleted the drives and I'm starting over. The drives seem to be ok, but under heavy load, they seem to disconnect from the bus.

I had similar issues with my Samsung 750GB drives that would disconnect under windows, but i was able to remedy that issue by disabling NCQ. Under OpenFiler I don't see any way of doing that with these 1.5TB drives. So, I'm kinda left thinking I'll ditch this solution for a new motherboard with a better SATA controller.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Seems freeNAS locks up pretty hard when I try to create a RAID 5 array also. Back to the drawing board I guess.
 

Stereodude

Not really a
Joined
Jan 22, 2002
Messages
10,865
Location
Michigan
Yup, that is the expected behavior. I've had arrays take a week to rebuild. The drive that is pegged is the one storing the CRC data, it will cycle through all the drives eventually.
Raid 5 doesn't have a drive that stores CRC / Parity data. The parity data is spread across all the drives. My RAID5/6 arrays on the Dell PERC cards built arrays with 6 or 8 1.5TB 5400RPM drives in less than 5 hours.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Which PERC cards did you get? I'm looking on eBay but I can't really figure out which one to get.

My other thought was to stay with the software raid and go with a newer board like this gigabyte GA-P55-UD5 which has 10 SATA ports on it.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
I'm looking on Dell's website. Did you get the integrated 2x4 internal with 256MB RAM? Did it need any special cables or did yours come with them? What OS are you using with the adapter?

Code:
PERC 6/I Integrated / Adapter 	3Gb/s SAS	PCI-Express 1.0	2x4 internal	256MB	Yes (BBU)	0,1,5,6,10,50,60	16	Hardware RAID
 

Stereodude

Not really a
Joined
Jan 22, 2002
Messages
10,865
Location
Michigan
Yes, that's it. I bought cables from ebay pretty cheap. I bought the cards from a forum member at overclock.net ~$100 each with battery backup and they did not come with cables. I'm not sure if the cards from Dell come with cables or not. I'd guess not.

I'm using XP x64.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Stereodude, does yours look like this:

341-9288.jpg


I can't seem to find the Perc 6 card that looks like a normal PCIe card.
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,719
Location
Horsens, Denmark
Raid 5 doesn't have a drive that stores CRC / Parity data.

That depends on how it implements RAID 5. The older approach involves each stripe being composed of n-1 drives in (essentially) RAID 0 with the last drive in each composing CRC info. The drive that stores CRC info does cycle from drive to drive for each stripe. This was the main performance bottleneck for larger arrays (the write speed on the single CRC drive). IIRC, more sophisticated controllers took to dividing up all the CRC information and writing a part of it on each drive along with the data, reducing the bottleneck, but at the cost of significant complication on the computation side.
 

Pradeep

Storage? I am Storage!
Joined
Jan 21, 2002
Messages
3,845
Location
Runny glass
Does it normally take a long time for a software based RAID 5 array to be built? I just setup OpenFiler with their software RAID using my five 1.5TB drives as a single RAID 5 volume. The array is busy building for the first time and it's been running since 6:00AM Sunday morning and it's only at 83% complete (~30 hours). I'm assuming it does take a while to initially build the array for the first time? The activity lights have an interesting pattern with 4 of the 5 drives blinking fast and one staying a bright solid...but it shifts around.

Not Seagates by any chance? There are several that really shit the bed when asked to work in a RAID 5 config. I don't think a new controller will help in that case.
 

Stereodude

Not really a
Joined
Jan 22, 2002
Messages
10,865
Location
Michigan
That depends on how it implements RAID 5. The older approach involves each stripe being composed of n-1 drives in (essentially) RAID 0 with the last drive in each composing CRC info. The drive that stores CRC info does cycle from drive to drive for each stripe. This was the main performance bottleneck for larger arrays (the write speed on the single CRC drive). IIRC, more sophisticated controllers took to dividing up all the CRC information and writing a part of it on each drive along with the data, reducing the bottleneck, but at the cost of significant complication on the computation side.
That would be RAID-3 or 4. RAID-5 has always had distributed parity as far as I'm aware.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
I had the same kind of moment the other day when I was trying to explain the difference between RAID 1 and RAID 10. I would have also failed Chewy's exam. However, it was worth it to refresh my understanding and explain in greater detail to someone else.
 
Last edited:

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Not Seagates by any chance? There are several that really shit the bed when asked to work in a RAID 5 config. I don't think a new controller will help in that case.

No, I bought the Samsung Ecogreen F2 1.5TB drives for this project. Are there some drives more suitable for this than othes? I read some recommend the TLER (or similar), but after reading about it, it didn't sound useful without a RAID controller. All five of my drives stay right around 25 degrees C because of active cooling from my Supermicro carrier.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
You just remove it from the carrier and put a normal PCI bracket on it. link

Thanks for the link. I also found a place that sells the 1 > 4 SATA cables for this card for around $5 each. I read in other places that active cooling is a must for this card. Are you doing anything special with yours? Did you change the heatsink on it or anything?
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,719
Location
Horsens, Denmark
Most server-class RAID cards expect very significant airflow around the card. I ended up replacing the tiny aluminum heatsink on my 3Ware 9650-16ML with a much larger copper one, and outright added a large aluminum heatsink to a chip that was bare otherwise. Even with these changes, I had to add a dedicated case fan or the temps went over 100F.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
My current AMD nForce 570 chipset doesn't support AHCI, so I can't use that feature.

Unrelated, I updated my motherboard's BIOS and also ran an update command inside OpenFiler and things seem to be running better right now.

I went through trying OpenFiler, FreeNAS, Windows Home server, Windows Storage server 2008, and Server 2008 R2 last night.

Windows Home server had weird problems with my hardware (even after updating drivers). I would transfer large files (4GB+) over to a share and 2/3rds of the way through the transfer things were fast (50-60MB/sec), but the transfer rates went down to extremely slow rates (<1MB/sec) in the last 1/3rd no matter what I tried.

Windows Storage Server just seemed too complicated for my needs. I couldn't get a RAID 5 software volume to work and gave up trying to figure out how. Every time I tried to add it, windows said that object wasn't supported.

Server 2008 R2 actually worked well. Transfer rates were reasonable and it support software RAID 5. I decided to try OpenFiler again after reading a bunch online about using the conary update even after downloading the latest build. They had at least 30+ updates available and the update mechanism was very slick. I also found that my BIOS was 7 revisions behind, so I decided to update it to see if that would help. So far it seems to be working ok, but the RAID 5 array is still rebuilding. Although it seems to be faster this time around with 55% complete after 12 hours.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,916
Location
USA
Yes, that's it. I bought cables from ebay pretty cheap. I bought the cards from a forum member at overclock.net ~$100 each with battery backup and they did not come with cables. I'm not sure if the cards from Dell come with cables or not. I'd guess not.

I'm using XP x64.

I know this isn't as cheap as you paid, but what do you think of this Perc 6i eBay listing?
 

Chewy509

Wotty wot wot.
Joined
Nov 8, 2006
Messages
3,348
Location
Gold Coast Hinterland, Australia
Yep.

So Dave, I guess this means you'd fail Chewy's job criteria (like the rest of us)?

Only if you said you were a RAID guru and you set them up on a regular basis... Knowing that RAID 0 is striped, RAID 1 is mirrored and RAID 5 is striped with parity is enough to earn a pass mark on that question.

BTW, riddle me this:

I have 5 weights, named M, N, O, P and Q. All 5 weights are different in weight, but I know that they weigh 1g, 2g, 3g, 4g, and 5g, but unsure which weight weighs what?

From 3 test with a balance, I find that:
1. O = M + N. (O is equal weight to that of M and N combined).
2. Q > O. (Q is heavier than O).
3. P < O. (P is lighter than O).

Can I determine what which weight weighs from the above information? And if not, what is the minimum number of additional tests needed to determine what each one weighs?

Free smiley to the first correct answer.
 

BingBangBop

Storage is cool
Joined
Nov 15, 2009
Messages
667
Answer:

m=1g
n=3g
o=4g
p=2g
q=5g

reason:

The choices that match 0 = m + n are
m n o
1 2 3
1 3 4
1 4 5
2 3 5
since q>o you can exclude all choices where o is 5
Since P<o you can exclude the choice where o=3 because 1 and 2 are already allocated.
that leaves m=1, n=3, o=4. because Q>o that means q has to be 5 and the only other number left is p=2.
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,719
Location
Horsens, Denmark
Lets see if serious sleep deprivation helps on this one...I'm guessing not.

With that information, we know that P=2, O=4, and Q=5. We still need to know whether M or N is heavier.
 

BingBangBop

Storage is cool
Joined
Nov 15, 2009
Messages
667
correction: there are more choices

m n o
1 2 3
1 3 4
1 4 5
2 3 5
2 1 3
3 1 4
4 1 5
3 2 5

The same exclusions apply as above and that gives two remaining choices

m n o p q
1 3 4 2 5
3 1 4 2 5

So one needs one more test. There are a variety of choices for that one test: m>n; n>m; m>p; n>p and the coresponding < tests
 

Chewy509

Wotty wot wot.
Joined
Nov 8, 2006
Messages
3,348
Location
Gold Coast Hinterland, Australia
The same exclusions apply as above and that gives two remaining choices

m n o p q
1 3 4 2 5
3 1 4 2 5

So one needs one more test. There are a variety of choices for that one test: m>n; n>m; m>p; n>p and the coresponding < tests

Correct. :D

Out of interest, how long did it take you to work it out?

(I asked this same question to most of the recent candidates we interviewed, and only 2 got the correct answer. I was looking for methodology in finding the answer, not the correct answer, and the most common approach was to guess the values, and then try the equations, rather than elimination of values).
 

BingBangBop

Storage is cool
Joined
Nov 15, 2009
Messages
667
It was quite quick, within a couple of minutes. I also noted my error instantaneously after posting the first answer. The critical thing is that i relied on paper and pencil to write down the o= m + n choices and that it made it visual. In an interview, with no pencil and paper avail and just using my memory with no visualization of the problem I don't think I would have solved it while under the gun either.

By the way ddrueding correctly solved it first! Not me, because of the error I made.
 

LunarMist

I can't believe I'm a Fixture
Joined
Feb 1, 2003
Messages
17,454
Location
USA
Which sort of jobs require that sort of mental masturbation? Are they entry level positions?
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,719
Location
Horsens, Denmark
I just did it in my head. I always got in trouble for not showing my work in school.

Thought process as I read it.

O is big
Q is bigger
P is smaller than both
Assume Q is the highest - the rest falls quickly into place. Under a minute.
 
Top