Ultimate Unraid Server - The Sequel

And you can't buy a 2TB Raptor, so it is a moot point for anyone who wants more than 300GB drives in unRAID. 

I agree that you can't buy a 2TB raptor, but you can buy a 2TB Caviar Black, 2TB RE4 and 2TB Barracuda XT.  All of these put in throughput numbers comparable to the Raptor (albeit at a higher latency). 

How about doing a Seagate 1.5TB mod to make it pretty much on par (and even faster in some cases) than a 300GB Velociraptor?

Interesting info, but I'm not sure what the point of this would be.  Performance after the mod is similar to the vRaptor but at a higher latency due to RPM and access time.  This hack brings the seagate drive capacity down to 300GB, the same as the vRaptor.

If you are writing 4GB, you need to take the time needed for the subject drive to read 4GB sequential, and add the time needed to write 4GB sequential.  Then divide 4GB by that total time, and you get the theoretical max write performance of unRAID for that drive.  Do the same for the parity drive.  Your max is the SLOWEST of the two.  If you can get over 90% of that max, you are doing good.

Good points.  I actually had calculated theoretical performance between the two.  I don't have the spreadsheet right now, but I believe it was 18% theoretical drop but we illustrated a 27% drop.

Overall, it's not that bad, considering the EARS are 54% the speed of the VRaptors on rotational matters. I think this indicates how performance is really dictated by multiple and not singular factors.

I agree - not bad at all considering the price (I've seen deal after deal on these drives lately - just posted another one this morning).

It does feel bad when waiting for the files to transfer, though. 

Those 4GB files take 3 minutes and 9 seconds to transfer on the high speed drives but take 4 minutes and 2 seconds on the EARS.  That extra minute of wait time feels like forever.  And since this is roughly the equivalent of a full DVD image transfer, I suspect this would be a typical scenario for a lot of people.

I believe this "entry level" system is only running on 1GB of memory in dual channel mode (I will have to confirm).  The Core2Quad test was a 4GB system. 

The i3-530 system I will start testing later this week will also be 4GB.

The Core2Quad test was a 4GB system. 

Then the whole things in meaningless unless you take steps to prevent buffering effects from swamping the dependent variables.

I moved the same pair of vRaptor baseline drives over to a new i3-530 H55 platform tonight.  

The board used is an ASUS P7H55-M PRO motherboard.  The board has 6 onboard SATA 3 ports and the processor has on package video.  Since I thought this would be a good blend of performance and power consumption, I also took power measurements below.

Same configuration as the other systems with the exception of the physical necessities of the platform (4GB DDR3 instead of 4GB DDR2).  I am still working on the write tests, but the read performance seems to scale quite a bit depending on the platform.

Note that I have not adjusted any unRAID parameters to attempt to optomize for the 4GB memory.  It is completely stock.

Power Consumption:

Pentium D 820, ICH7 (My 5 year old unRAID)

86W Idle (Disks Spun Down)

110W Idle (Disks Spun Up)

132W Peak During Active File Write

Atom 330, ICH7:

38W Idle (Disks Spun Down)

47W Idle (Disks Spun Up)

56W Peak During Active File Write

Core2 Quad Q8300, ICH10:

40W Idle (Disks Spun Down)

48W Idle (Disks Spun Up)

81W Peak During Active File Write

i3-530, H55

36W Idle (Disks Spun Down)

44W Idle (Disks Spun Up)

55W Peak During Active File Write

I'm blown away by the power consumption numbers on the i3-530.  I knew it would be close, but I am shocked that it actually beats the overall power consumption of the Atom 330 ICH7 with onboard video.  I haven't actually timed it, but unRAID is operational in less than 30 seconds after power on with the i3 while the Atom system chugs and takes several minutes.  The read performance on the identical drives is up 55% from the numbers posted by the Atom based board.  I haven't finished the write performance test, but early numbers look very similar to the results of the Atom and Core2Quad posted earlier.

The read performance on the identical drives is up 55% from the numbers posted by the Atom based board.

As would be expected as the Atom has an ICH7... you are really comparing chipsets and mobos, not CPUs.

And you are getting huge buffer effects... you have to account for buffers by tying up 90% of RAM to prevent it from being used by buffers.

The read performance on the identical drives is up 55% from the numbers posted by the Atom based board.

 

As would be expected as the Atom has an ICH7... you are really comparing chipsets and mobos, not CPUs.

 

And you are getting huge buffer effects... you have to account for buffers by tying up 90% of RAM to prevent it from being used by buffers.

It seems there is more going on here than just the chipset.  I don't see how to explain the similar boost in read capacity from the Core2 Q8300 to the i3-530 on H55.  My understanding is the ICH10 architecture is comparable to the H55.  

Also:  with the power draw where it is, why would a person not favor the additional processing potential of a core i3?  Based on the results so far, I am leaning heavily toward this platform for the blend of performance and efficiency I have seen.  However, I'm certainly not finished with evaluating everything yet.  

In the another thread that is discussing the new Supermicro X7SPA ATOM pinetrail board (I believe it uses an ICH9R controller), the power numbers posted are actually worse than these.  However, I believe the measuring system is suspect.  I have been trying to get someone to take better measurements on it, but no takers so far...

I received a second WD10EARS 1TB drive yesterday and installed both in the i3-530 system (my current "best case" system).  The system now has a matched pair for Parity and Data.

The read performance was outstanding.  Read throughput actually EXCEEDED the (2) vRaptors.  However, write performance declined as expected.  I will post the comparative numbers tonight. 

EDIT:  Reran this test and the read performance was not as high as reported last night.  Still not bad:

Parity:  WD10EARS, Data:  WD10EARS

Write:  28% slower than vRaptors 

Read:  9% slower than vRaptors

As I get more data together, I believe I will keep two charts (they are growing crowded):

1) The best overall drive performance moving between multiple system architectures/chipsets (same memory amount)

2) Various hard drive configurations on the highest performing system architecture.

I would love to get a pair of 2TB WD RE4 drives into the test, but they are cost prohibative right now at over $300/ea.

Also:  with the power draw where it is, why would a person not favor the additional processing potential of a core i3?  Based on the results so far, I am leaning heavily toward this platform for the blend of performance and efficiency I have seen.  However, I'm certainly not finished with evaluating everything yet.

Well, if your system is up and running 24/7 with all drives spun up all the time, then, maybe, power draw is not _that_ much different, but if you think about unRaid being in idle state most of the time, then it makes much bigger difference. My 2c.

Well, if your system is up and running 24/7 with all drives spun up all the time, then, maybe, power draw is not _that_ much different, but if you think about unRaid being in idle state most of the time, then it makes much bigger difference. My 2c.

The new Supermicro pinetrail based ATOM system ended up using 9 watts less than the i3-530 at idle.  That's only $8.67 a year if you pay $.11 per kWH.  

I ended up purchasing the i3-530 after I completed researching the performance aspects, but the Supermicro is a solid performer with a great feature set and excellent power efficiency.  Cost is about $200 either way.  If I end up doing something different with my system 6 months from now, the i3-530 system will be more useful to me.

The new Supermicro pinetrail based ATOM system ended up using 9 watts less than the i3-530 at idle.  That's only $8.67 a year if you pay $.11 per kWH. 

 

I ended up purchasing the i3-530 after I completed researching the performance aspects, but the Supermicro is a solid performer with a great feature set and excellent power efficiency.  Cost is about $200 either way.  If I end up doing something different with my system 6 months from now, the i3-530 system will be more useful to me.

There is no question about the performance and possibility to expand. I totally agree.

In my case electricity is somewhat more expensive (~40c per kWh), so I've been thinking about Atom (Supermicro X7SP) as a possible solution, but I keep thinking about i3 as well and so far couldn't make up my mind - just researching the subject. I'm surprised to see your results for power consumption.

Also, reading this thread I keep wondering as to why are you so concerned about read/write performance? If you are looking to find a best performing system, then wouldn't it be wise to look at hardware RAID solutions with high performance HDDs as oppose to software solution like unRaid? Or is this just to have faster parity checks? I don't mean this as a disrespect or anything negative, I just have hard time understanding the reason behind all this except for the sake of it. It's the same to me as when people are talking about getting 7200 rpm drives as oppose to lower power 'green' drives. Won't network saturate much faster which will negate the performance of the drives anyway?

In my case electricity is somewhat more expensive (~40c per kWh), so I've been thinking about Atom (Supermicro X7SP) as a possible solution, but I keep thinking about i3 as well and so far couldn't make up my mind - just researching the subject. I'm surprised to see your results for power consumption.

 

Also, reading this thread I keep wondering as to why are you so concerned about read/write performance? If you are looking to find a best performing system, then wouldn't it be wise to look at hardware RAID solutions with high performance HDDs as oppose to software solution like unRaid? Or is this just to have faster parity checks? I don't mean this as a disrespect or anything negative, I just have hard time understanding the reason behind all this except for the sake of it. It's the same to me as when people are talking about getting 7200 rpm drives as oppose to lower power 'green' drives. Won't network saturate much faster which will negate the performance of the drives anyway?

Wow, that's some rediculously high electrical billing.  The atom makes a good platform here.  I personally hate ATOM in netbooks, but it works well for me in various web servers and the new supermicro board has a fantastic feature set.  I am a long time user of unRAID and just like it.  I know full well there are higher performance options out there, but I simply want to get the most out of the system that I have used for so long.  

Green drives perform very well (as long as you make sure the new EARS are alligned to the 4k sectors), but even in unRAID there can be a very measurable performance penalty vs high performance drives.  In some cases, I saw a 38% improvement in write performance when using high performance drives (110MB/s) vs WD EARS green drives (85MB/s).  Of course this only amounted to 16.7MB/s vs 21.6MB/s performance.  Neither are stellar vs other hardware RAID NAS configurations.  Read performance for the green drives was only off by 4.1% (42.9MB/s vs 44.7MB/s).

The chipset/platform made a bigger difference for me.  Again, the big difference was seen in write performance where my i3-530 with H55 chipset performed at 44.7MB/s vs 29.5MB/s read rates for the same hard drives in an Atom 330 ICH7 chipset system.  To be fair, however, I believe the biggeset limitation was the chipset rather than the processor.  The supermicro uses the ICH9R which is a much better option.  

However, I am very happy with the gains I have made with my upgrade.  At the beginning of this post, many people stated that my performance would not change and that my old system was just as good.  For me, these improvements were substantial and well worth my cost to move to the i3-530 and H55 chipset:  I reduced my idle power consumption by 50 Watts while at the same time increasing my transfer rate by 42% (Read) and 44% (write) while using the same 7200 RPM hard drives I used in the old system!  

The power savings alone will pay for the cost of the upgrade in 4 years!

 

If you have a better suggestion to compare performance, please advise.  If you have nothing of value to contribute, please move along.

 

 

Interesting.

 

Just what value have you contributed?

 

 

 

Wow, this is turning into a flame war very quickly.  The dramallama has no place here.

 

GaryMaster: Please just listen to what the experienced unRAID users in this thread have already said, especially the Hero members.  They know what they are talking about.  For the last time, the benchmarks you are asking for are useless and don't apply to the unRAID environment.  If you want speed over data protection and ease-of-use, then unRAID is not for you.

 

Everyone else: Keep in mind that GaryMaster is clearly new to unRAID and comes from the world of performance-at-all-costs.  He is used to the CPU and RAM being critical in any situation.  It took me a while to get used to this concept when I first discovered unRAID as well (though, admittedly, I didn't get belligerent over the idea!).  If he had a front wheel drive sports car, he would probably want to perform benchmarks on various trunk-mounted spoilers. (sorry, I just had to work that in there somewhere )

 

I vote this thread be closed/locked.  There is nothing productive that is going to come of this.

Wow, thanks for saying this Rajahal . I have not been on the forums for a little while.

I was surprised to see how people were interacting all of a sudden, in a non friendly way.

Now some of us do more with unraid then just a NAS (sabnazbd, VMware, and so on) and then the extra cpu makes a difference. But for reading files off the unraid server for 1080p movies, you do not need much horse power at all. I actually like the post about using an atom pc to do the work.

Personally I have built many systems of unraid now for both myself, friends and colleagues, with different levels of chipsets, hard drives, cpu. I have seen performance gains (5MB/s), but only slight and not worth the investment. I even had an onboard Raid 0 working with unraid, (search for my posts with benchmarks), but in the end I preferred to not go that route, as I was getting errors with my drive not able to power down. I prefer to be 100% compatible and not worry about data loss.

Instead of worrying about how fast you can write to the box, do your work on the unraid box so you do not have to transfer your files. If you are downloading, then do it on the unraid box directly. I recommend downloading to your cache drive directly, then moving only what you want to keep. That is how I do it now, and it is much faster. Also doing it this way, you do not need to leave a second computer on all night or day. GaryMaster, why do you want the speed so badly? Just for shits and giggles or do you have a need to fill?

What we really should have is a list of tested GB switches and routers, Networks cards, settings. I have yet to break the 108MB/s ram to ram transfer speed between two unix boxes on 1GB, no jumbo frames, Cicso managed switch though. Has anyone tested 10G? Prices will eventually come down. I understand GaryMaster's wanting more speed. It is like a drug! In the boating world we call it 1 foot envie. We always want the boat that is 1 fot bigger, 1 Knot faster...

Folks keep it friendly, that is how we attract more users and make a better product.

Wow, that spark was buried 2 months ago and you wanted to dig it up?  

Lets see if this can start a new flame.

That's kind of what I thought... I was sort of trying to forget about that bad start that we got off to and wrap this up with the summary of where I ended up in post #187.  Here... I'll try again:

In my case electricity is somewhat more expensive (~40c per kWh), so I've been thinking about Atom (Supermicro X7SP) as a possible solution, but I keep thinking about i3 as well and so far couldn't make up my mind - just researching the subject. I'm surprised to see your results for power consumption.

 

Also, reading this thread I keep wondering as to why are you so concerned about read/write performance? If you are looking to find a best performing system, then wouldn't it be wise to look at hardware RAID solutions with high performance HDDs as oppose to software solution like unRaid? Or is this just to have faster parity checks? I don't mean this as a disrespect or anything negative, I just have hard time understanding the reason behind all this except for the sake of it. It's the same to me as when people are talking about getting 7200 rpm drives as oppose to lower power 'green' drives. Won't network saturate much faster which will negate the performance of the drives anyway?

Wow, that's some rediculously high electrical billing.  The atom makes a good platform here.  I personally hate ATOM in netbooks, but it works well for me in various web servers and the new supermicro board has a fantastic feature set.  I am a long time user of unRAID and just like it.  I know full well there are higher performance options out there, but I simply want to get the most out of the system that I have used for so long.  

Green drives perform very well (as long as you make sure the new EARS are alligned to the 4k sectors), but even in unRAID there can be a very measurable performance penalty vs high performance drives.  In some cases, I saw a 38% improvement in write performance when using high performance drives (110MB/s) vs WD EARS green drives (85MB/s).  Of course this only amounted to 16.7MB/s vs 21.6MB/s performance.  Neither are stellar vs other hardware RAID NAS configurations.  Read performance for the green drives was only off by 4.1% (42.9MB/s vs 44.7MB/s).

The chipset/platform made a bigger difference for me.  Again, the big difference was seen in write performance where my i3-530 with H55 chipset performed at 44.7MB/s vs 29.5MB/s read rates for the same hard drives in an Atom 330 ICH7 chipset system.  To be fair, however, I believe the biggeset limitation was the chipset rather than the processor.  The supermicro uses the ICH9R which is a much better option.  

However, I am very happy with the gains I have made with my upgrade.  At the beginning of this post, many people stated that my performance would not change and that my old system was just as good.  For me, these improvements were substantial and well worth my cost to move to the i3-530 and H55 chipset:  I reduced my idle power consumption by 50 Watts while at the same time increasing my transfer rate by 42% (Read) and 44% (write) while using the same 7200 RPM hard drives I used in the old system!  

The power savings alone will pay for the cost of the upgrade in 4 years!

[/END THREAD]

Wow, that spark was buried 2 months ago and you wanted to dig it up? 

Lets see if this can start a new flame.

Smino got it all wrong there. 

Now we all love each other very much.

Smino got it all wrong there. 

Now we all love each other very much.

Yea... some of us were actually getting together this weekend to sing campfire songs and do some trust falls.

We all have different needs.

I for one learned a little in this thread.

I was going to post something the other day regarding spare cpu power, but now it has reminded me.

I found that having the cpu power to spare comes into play for certain validation operations.

In doing a directory sweep of md5sum for my files, I found the extra horsepower helped after comparing the run times between the source machine and destination machine (unraid).

I was actually considering a switch to the atom processor, but after that exercise, I thought it may be better to move to a higher processor with lower power utilization.

In the future I plan to build some form of directory sweep validation mechanism which runs in cron.

I think this actually may help for those who make minor errors on their systems.

Using par2 we may even be able to reconstruct corrupted parts.

I mention this because, unraid IS my backup solution and any form of protection can go a long way.

So having the extra processing power may pay in the end (for me).

although I'm not sure of the thread title heh    I found some of this thread interesting.

We all have different needs.

I for one learned a little in this thread.

I was going to post something the other day regarding spare cpu power, but now it has reminded me.

 

I found that having the cpu power to spare comes into play for certain validation operations.

In doing a directory sweep of md5sum for my files, I found the extra horsepower helped after comparing the run times between the source machine and destination machine (unraid).

 

I was actually considering a switch to the atom processor, but after that exercise, I thought it may be better to move to a higher processor with lower power utilization.

 

In the future I plan to build some form of directory sweep validation mechanism which runs in cron.

I think this actually may help for those who make minor errors on their systems.

 

Using par2 we may even be able to reconstruct corrupted parts.

 

I mention this because, unraid IS my backup solution and any form of protection can go a long way.

So having the extra processing power may pay in the end (for me).

 

although I'm not sure of the thread title heh    I found some of this thread interesting.

I like where your going with the par2 idea. For me I have offsite backups, but I wish there was a way to test existing music files for corruption (mp3, flac, ape, apple lossless...) Before unraid I was on a raid 5 software windows 2003 server. When it went down, most of my ape files became corrupt. ARGH! Oddly enough the mp3, flac, and apple lossless survived.

Had I had par2 files for them, I could of saved myself some time. How much added space do your par2 files take up, and where do you store your par 2 files? I would not want them in the same directory as the files movies/music themselves.

I am actually thinking of setting up an atom based pc for offsite Rsync type setup. Not sure if I will use unraid off site, although I do have a second lic.

I like where your going with the par2 idea. For me I have offsite backups, but I wish there was a way to test existing music files for corruption (mp3, flac, ape, apple lossless...) Before unraid I was on a raid 5 software windows 2003 server. When it went down, most of my ape files became corrupt. ARGH! Oddly enough the mp3, flac, and apple lossless survived.

Had I had par2 files for them, I could of saved myself some time. How much added space do your par2 files take up, and where do you store your par 2 files? I would not want them in the same directory as the files movies/music themselves.

I am actually thinking of setting up an atom based pc for offsite Rsync type setup. Not sure if I will use unraid off site, although I do have a second lic.

I have not fully thought this out, but I was considering using a separate disk or directory as a par2 database. It would recreate the same directory structure as the source. The only files in the directory would be the par2 files.

For md5sum, I was considering a gdbm database, full path as key, md5sum as data.

This could also be done as sqlite.

There are some patches for bash that put this functionality directly into the shell.

In any case, this is where the extra cpu can be a help.

Hello, I am a new unRAID user, and I would just like to point out something if it has been missed.

Practically the mohterboard listed in the OP, the Gigabyte Corei7 board will only have a use of 6 onboard sata ports. as the 4 white ones are hardcoded as raid 0/1 and you cannot reflash the port multiplier for it to work as seperate drives.

I get 40MB/sec write and 60MB/sec read off a single disk. I an not sure if this is any good by standards here on these fourms, but I am happy with this level of performance.

Hello, I am a new unRAID user, and I would just like to point out something if it has been missed.

 

Practically the mohterboard listed in the OP, the Gigabyte Corei7 board will only have a use of 6 onboard sata ports. as the 4 white ones are hardcoded as raid 0/1 and you cannot reflash the port multiplier for it to work as seperate drives.

 

I get 40MB/sec write and 60MB/sec read off a single disk. I an not sure if this is any good by standards here on these fourms, but I am happy with this level of performance.

What drives do you have attached?

I am planning to build the fastest UNRAID server possible

Gary, it looks like you have some serious competition:

I'm looking to build a very high-performance unraid server

 

Suddenly, I feel like I'm in a bar with some guy saying "Hey, that guy over there says he thinks you're a wuss!".

Suddenly, I feel like I'm in a bar with some guy saying "Hey, that guy over there says he thinks you're a wuss!".

Yep! 

Oh, hold on... let me grab my popcorn...  

I don't know, a 'very high performance system' is a far cry from 'the ultimate unRAID server'...