What do you think of this HBA card?

 

http://www.newegg.com/Product/Product.aspx?Item=N82E16816118142

 

It's a bit cheaper but has 16 ports. So you need to have 6 ports on your motherboard if you want use all 24 bays.

 

For starters, it only has 1/4th of the bandwidth of the 2760A 

 

If you're talking PCIe bandwidth, it's not that bad.  With 16 drives, each drive would see 250MB/s of bandwidth.  The 2760A, with 24 drives, has 333MB/s.  So, technically, it has 3/4 the bandwidth of the 2760A.  Remember, the 2760A is PCIe 2.0 only, not 3.0.  I'm sure the PCIe 3.0 version will be announced real soon, now that I've placed my order...    I think 250MB/s is a good number.

 

That said, I'm not sure what chips are in play on the board itself.  The 2760A is special in that each drive get's the full SATA bandwidth spec.  If this LSI card is using port multiplier technology, then each drive would have less than full SATA bandwidth spec.  So even though the PCIe bus connection gives plenty of bandwidth for each drive, the card itself may not be the highest performer (but probably not bad either, just average).

 

By the way, MvL, you're numbers are a touch off.  You would need 8 on-board SATA ports to get up to 24 total with this card. 

 

Pricewise, yes it is cheaper, but not cheaper per port.  $409/16=$25/port, $542/24=$23/port. 

 

Also consider the motherboard I just picked up was only $50 (yes, it only had 2 SATA ports, but I don't need any motherboard ports).  I only found 1 socket 1155 motherboard that had at least 8 SATA3 ports, and it was $400 (ouch!).  The cheapest motherboard with 6 SATA3 ports (with enough SATA2 ports to get you to 8 total) was $150.  If you drop down to 4 SATA3 and 4 SATA2 ports, you can find options in the $75 range, but they don't look like good choices for motherboards - I think you would be back up around $100 for a decent option.

 

All said, the price premium for the 2760A is not much when you start running the numbers.  And assuming it lives up to the performance potential, I don't think anything can touch it for the money.

 

Of course, we are dealing with 5400rpm drives here, so chances are none of this will amount to much. 

You're right -- I had forgotten the 2760A was only PCIe v2 

 

Nevertheless, if you were going to buy a 16-port card, I'd just go with the RocketRAID 2740, which is actually cheaper than the card MvL listed ($383.76 at Provantage instead of $409 ... or $23.99/port)  and is a PCIe x16 card using the same switching technology as the 2760A [it's simply the "little brother" of the 2760A  ]    In fact, as you just implicitly pointed out, with the PCIe v2 bandwidth limits the 2740 would actually outperform the 2760A if you were using an array of SATA-3 SSDs (since it would have 500MB/drive of available bandwidth) !!

(Not a likely scenario)

 

Bottom line:  Your summary is right on !! ["All said, the price premium for the 2760A is not much when you start running the numbers.  And assuming it lives up to the performance potential, I don't think anything can touch it for the money."]

 

 

I've thought about using a PicoPSU on my Atom build for the same reason.  I suspect I'm wasting at least 5W just due to power supply efficiency (or, more accurately, inefficiency) at such low loads.    I didn't do it just because of concern about spinup draw ... but I may have to try that just for grins 

 

I'm considering using one as well. The 160W PicoPSU would definitely be sufficient to power my build at max load (parity check), the concern is drive spinup. With 5 HDD, the initial power on will be real close to max output of the PSU. As long as it can successfully boot up it should be fine as it would only have to spin up 4 drives to run a parity check as the cache drive never spins down.

 

If it did work, it wouldn't leave any overhead for additional drives, but I don't see myself adding any more drives any time soon.  And it would certainly free up a lot of room in that cramped little case!

  Ok, the chip on the board, Marvell 88SE9485, is rated to consume 5.5 watts, so with 4 of them, expect over 22 watts, plus the power of other components on the board... http://www.marvell.com/storage/system-solutions/assets/Marvell-88SE9485-9445-Product-Brief.pdf

I've thought about using a PicoPSU on my Atom build for the same reason.  I suspect I'm wasting at least 5W just due to power supply efficiency (or, more accurately, inefficiency) at such low loads.    I didn't do it just because of concern about spinup draw ... but I may have to try that just for grins 

 

I'm considering using one as well. The 160W PicoPSU would definitely be sufficient to power my build at max load (parity check), the concern is drive spinup. With 5 HDD, the initial power on will be real close to max output of the PSU. As long as it can successfully boot up it should be fine as it would only have to spin up 4 drives to run a parity check as the cache drive never spins down.

 

If it did work, it wouldn't leave any overhead for additional drives, but I don't see myself adding any more drives any time soon.  And it would certainly free up a lot of room in that cramped little case!

 

r.e. "free up a lot of room in that cramped little case" ==> You can do that with an SFX unit like this:  http://www.newegg.com/Product/Product.aspx?Item=N82E16817151063    But obviously the PicoPSU would free up a LOT of room [i.e. the entire PSU bay  ]

 

But I agree the PicoPSU is VERY tempting => no minimum load requirement; over 90% efficiency; and supports a peak load of 15A (180W) on the 12v line, so it shouldn't have any problem spinning up my 6 WD Reds (~ 20W spinup current).

 

I could probably get my idle consumption down to ~ 18W or so with that PSU !!

 

  Ok, the chip on the board, Marvell 88SE9485, is rated to consume 5.5 watts, so with 4 of them, expect over 22 watts, plus the power of other components on the board... http://www.marvell.com/storage/system-solutions/assets/Marvell-88SE9485-9445-Product-Brief.pdf

 

I suspect that rated power draw is at load.  Since this board is going to be used solely for the SATA interface -- NOT with any of the RAID functions in use -- I'd expect the power draw to be somewhat less than that.    Just how much less is obviously an unknown ... depends on the design of the chip.

  A little curiosity I noticed about power supplies...  The efficiency also varies depending on the supply voltage used with the power supply, at least on the ones I have looked into in detail.  It seems most, if not all multi-voltage input switching power supplies are also more efficient at 240 VAC input voltage than they are at the US standard of 120 VAC!

 

  Since in the USA we also have 240 VAC available for appliances in most homes and buildings, I was thinking of running separate 240 VAC branch circuits to my computers next...  (after all I already have done that for my imported game consoles...)

 

  A little curiosity I noticed about power supplies...  The efficiency also varies depending on the supply voltage used with the power supply, at least on the ones I have looked into in detail.  It seems most, if not all multi-voltage input switching power supplies are also more efficient at 240 VAC input voltage than they are at the US standard of 120 VAC!

 

  Since in the USA we also have 240 VAC available for appliances in most homes and buildings, I was thinking of running separate 240 VAC branch circuits to my computers next...  (after all I already have done that for my imported game consoles...)

 

Yes, 240v is more efficient ... in fact, the higher the voltage, the higher the efficiency (all other things being equal).    That's why major appliances are all 240v; many high-end electrical tools are 480v;  and why power transmission lines are MUCH higher (e.g. 110,000v and even higher).

 

r.e. "free up a lot of room in that cramped little case" ==> You can do that with an SFX unit like this:  http://www.newegg.com/Product/Product.aspx?Item=N82E16817151063    But obviously the PicoPSU would free up a LOT of room [i.e. the entire PSU bay  ]

 

But I agree the PicoPSU is VERY tempting => no minimum load requirement; over 90% efficiency; and supports a peak load of 15A (180W) on the 12v line, so it shouldn't have any problem spinning up my 6 WD Reds (~ 20W spinup current).

 

I could probably get my idle consumption down to ~ 18W or so with that PSU !!

 

Yeah, but I'd love to free up the entire PSU bay, plus it would eliminate a source of noise (PSU fan) as well as heat (by moving the actual AC/DC converter outside of the case).  I'll probably give it a shot here soon.

 

I could probably lower my idle consumption quite a bit as well, as at idle my current PSU is at roughly 7.5% load, which equates to very low 70's efficiency on my particular PSU from tests I've seen.  I could probably lower my idle to ~25W with a PicoPSU.

I've thought about using a PicoPSU on my Atom build for the same reason.  I suspect I'm wasting at least 5W just due to power supply efficiency (or, more accurately, inefficiency) at such low loads.    I didn't do it just because of concern about spinup draw ... but I may have to try that just for grins 

 

I'm considering using one as well. The 160W PicoPSU would definitely be sufficient to power my build at max load (parity check), the concern is drive spinup. With 5 HDD, the initial power on will be real close to max output of the PSU. As long as it can successfully boot up it should be fine as it would only have to spin up 4 drives to run a parity check as the cache drive never spins down.

 

If it did work, it wouldn't leave any overhead for additional drives, but I don't see myself adding any more drives any time soon.  And it would certainly free up a lot of room in that cramped little case!

 

r.e. "free up a lot of room in that cramped little case" ==> You can do that with an SFX unit like this:  http://www.newegg.com/Product/Product.aspx?Item=N82E16817151063    But obviously the PicoPSU would free up a LOT of room [i.e. the entire PSU bay  ]

 

But I agree the PicoPSU is VERY tempting => no minimum load requirement; over 90% efficiency; and supports a peak load of 15A (180W) on the 12v line, so it shouldn't have any problem spinning up my 6 WD Reds (~ 20W spinup current).

 

I could probably get my idle consumption down to ~ 18W or so with that PSU !!

 

I did a bit of research yesterday trying to determine the maximum number of drives the PicoPSU 160 could spin-up.  I never found a solid answer, but the general consensus was about 5 3.5" drives.  I don't think the full 15A is available for spinning drives once you're powering a PC with it too.  I'd be very interested in some real results if one of you take the plunge.

 

I'm also getting conflicting information about how efficient these PicoPSU's really are.  The PicoPSU takes 12V power in (from a power brick), and converts some of it to 5V & 3.3V for the PC, and passes some unconverted as 12V.  The PicoPSU's conversion from 12V to 5V/3.3V is very efficient, 94%+.  But some people are saying that the power brick itself, which converts the household AC to 12VDC, is only about 70% efficient.

 

This is very confusing to me, as the PicoPSU power draw numbers seem to be 5W-10W below what normal power supplies consume.  I don't know what to believe.  I've never really read anything negative about the PicoPSU's, so perhaps that 70% number is made up.  Not knowing has caused me to shy away from pursuing a home-made power supply utilizing PicoPSU's plus extra power adapters for the hard drives, because that would be a total waste of time if the power bricks are really that inefficient.

 

Regardless, I picked up a KingWin LZP-650, which is a 650W Platinum power supply.  It has excellent low power efficiency in the neighborhood of 40W to 45W - only about 7W lost as heat, making it 80%+ efficient at ~6% load.  This is significantly better than most power supplies of this size.  It was a pricey $170, making it more expensive than the processor, motherboard and memory combined!

 

I did some unRAID spin-up testing yesterday, and everyone was right, unRAID starts all drives simultaneously.  Shameful.  That's why I had to get the 650W beastie instead of a 300W petitie.

 

By the way, garycase I'm a little confused by your build.  You mentioned you had a 20TB server, made with 6 Red 3TB drives.  At first I thought you had just rounded up a bit, from 18TB to 20TB, and that you really had a 7th drive for parity that you hadn't mentioned (since your case has room for a 7th drive internally).  Now that you're talking about spin-up for only 6 drives, the math tells me you can't have more than 15TB if one is for parity.  Was the 20TB just a typo?

  A little curiosity I noticed about power supplies...  The efficiency also varies depending on the supply voltage used with the power supply, at least on the ones I have looked into in detail.  It seems most, if not all multi-voltage input switching power supplies are also more efficient at 240 VAC input voltage than they are at the US standard of 120 VAC!

 

  Since in the USA we also have 240 VAC available for appliances in most homes and buildings, I was thinking of running separate 240 VAC branch circuits to my computers next...  (after all I already have done that for my imported game consoles...)

 

Very interesting.  I had noticed the same, but it never occurred to me to go that route.  What about the other components, like monitors/TV's?

 

Since I make a living with computers and monitors (and lot's of them), and I work from my home office, this might be worth pursuing for me.  I wouldn't retrofit my current home, but I'm planning to have a new home built in a couple years, and I could easily have 240V outlets installed for cheap in a few key areas.

 

Would I (or should I) have European 240V outlets installed?  Or would I have US 240V outlets?  Which is easier to plug into using the components we're talking about?

 

Would it even pass inspection if I had European outlets installed?  Hmmm...

 

I'm also getting conflicting information about how efficient these PicoPSU's really are.  The PicoPSU takes 12V power in (from a power brick), and converts some of it to 5V & 3.3V for the PC, and passes some unconverted as 12V.  The PicoPSU's conversion from 12V to 5V/3.3V is very efficient, 94%+.  But some people are saying that the power brick itself, which converts the household AC to 12VDC, is only about 70% efficient.

 

It's true that while the PicoPSU itself is super efficient, the AC/DC power brick is less so.  That said, you can find power bricks that are much more efficient than standard ATX PSU's.  The power brick that is sold on mini-box's website in conjunction with a PicoPSU is listed as >87% efficient. 

 

The other part of the equation is when you use a power brick, it has a much lower max power draw than most ATX PSU's, so you're running the brick at a higher % load which gives better efficiency.  Basically running the power brick at say 30% load, instead of an ATX PSU at <10% load.  And as we all know, ATX PSU's running at less than 20% load are horribly inefficient.

 

Most tests I have seen of PicoPSU/power brick combos result in >=80% efficiency, even at low loadings (where most ATX PSU's get around 70%), and as high as 87% efficiency at moderate loadings.  Of course, if you can find a low enough wattage ATX PSU to keep you at >=20% load at idle (while still having enough headroom for the peaks associated with spinup) you'll get 80%+ efficiency that way, easier and cheaper.

By the way, garycase I'm a little confused by your build.  You mentioned you had a 20TB server, made with 6 Red 3TB drives.  ... Was the 20TB just a typo?

 

Whoops!  What I MEANT to say was that YOU could build a 20TB server with that case/motherboard if you used 4TB drives.    I'm actually using 3TB drives, so it only has 15TB of capacity (I corrected that post).

 

Interesting topic!

 

Why not use the Celeron 1610T (low power model)? It has a TDP of 35 watts.

 

Good question.  Both processors idle the same.  The 1610T is capped on the high end (lower max frequency) to keep heat in check.  This allows you to use the processor in smaller cases, with smaller fans, and with smaller power supplies.  I wasn't worried about any of those issues.

 

Since my only concern was idle power, I save the money and went with the regular 1610, which will also have a little extra performance.

 

Agree.  There's virtually no reason to use the low-power versions of the i-series chips.  As Pauven noted, the idle power is virtually identical ... it's only the max TDP that's "throttled" on the low-power chips.  As long as you have ample power and cooling available, there's no reason to lose the extra "horsepower".    In fact, I replaced a Core i5-2400S I had used in one of my HTPCs with an i7-2700K to have better transcoding performance; and when I did that there was NO difference in the idle power consumption ... but significantly better processing capability (and of course notably higher power consumption under load).

 

Okay, didn't know that there was no difference in idle power consumption. Good to know!

What do you think of this HBA card?

 

http://www.newegg.com/Product/Product.aspx?Item=N82E16816118142

 

It's a bit cheaper but has 16 ports. So you need to have 6 ports on your motherboard if you want use all 24 bays.

 

For starters, it only has 1/4th of the bandwidth of the 2760A 

 

If you're talking PCIe bandwidth, it's not that bad.  With 16 drives, each drive would see 250MB/s of bandwidth.  The 2760A, with 24 drives, has 333MB/s.  So, technically, it has 3/4 the bandwidth of the 2760A.  Remember, the 2760A is PCIe 2.0 only, not 3.0.  I'm sure the PCIe 3.0 version will be announced real soon, now that I've placed my order...    I think 250MB/s is a good number.

 

That said, I'm not sure what chips are in play on the board itself.  The 2760A is special in that each drive get's the full SATA bandwidth spec.  If this LSI card is using port multiplier technology, then each drive would have less than full SATA bandwidth spec.  So even though the PCIe bus connection gives plenty of bandwidth for each drive, the card itself may not be the highest performer (but probably not bad either, just average).

 

By the way, MvL, you're numbers are a touch off.  You would need 8 on-board SATA ports to get up to 24 total with this card. 

 

Pricewise, yes it is cheaper, but not cheaper per port.  $409/16=$25/port, $542/24=$23/port. 

 

Also consider the motherboard I just picked up was only $50 (yes, it only had 2 SATA ports, but I don't need any motherboard ports).  I only found 1 socket 1155 motherboard that had at least 8 SATA3 ports, and it was $400 (ouch!).  The cheapest motherboard with 6 SATA3 ports (with enough SATA2 ports to get you to 8 total) was $150.  If you drop down to 4 SATA3 and 4 SATA2 ports, you can find options in the $75 range, but they don't look like good choices for motherboards - I think you would be back up around $100 for a decent option.

 

All said, the price premium for the 2760A is not much when you start running the numbers.  And assuming it lives up to the performance potential, I don't think anything can touch it for the money.

 

Of course, we are dealing with 5400rpm drives here, so chances are none of this will amount to much. 

 

True, I was late when i did the calculation. 

 

Out of interested do you need that speed of the 2760A? When buying this card it has RAID functionality is not better to use that then Unraid?

Interesting topic!

 

Why not use the Celeron 1610T (low power model)? It has a TDP of 35 watts.

 

Good question.  Both processors idle the same.  The 1610T is capped on the high end (lower max frequency) to keep heat in check.  This allows you to use the processor in smaller cases, with smaller fans, and with smaller power supplies.  I wasn't worried about any of those issues.

 

Since my only concern was idle power, I save the money and went with the regular 1610, which will also have a little extra performance.

 

Agree.  There's virtually no reason to use the low-power versions of the i-series chips.  As Pauven noted, the idle power is virtually identical ... it's only the max TDP that's "throttled" on the low-power chips.  As long as you have ample power and cooling available, there's no reason to lose the extra "horsepower".    In fact, I replaced a Core i5-2400S I had used in one of my HTPCs with an i7-2700K to have better transcoding performance; and when I did that there was NO difference in the idle power consumption ... but significantly better processing capability (and of course notably higher power consumption under load).

 

Okay, didn't know that there was no difference in idle power consumption. Good to know!

 

Yes, only reason to use a T series CPU is to keep heat in check.  I have an i3-3220T in my SFF build and went with the T series CPU specifically to limit the amount of heat generated in such a small case.  In addition to that, in the particular case I used, there's not a lot of clearance between the mobo and the PSU, so you need a relatively small CPU heatsink, which the T series CPU's have.

I'm also getting conflicting information about how efficient these PicoPSU's really are.  The PicoPSU takes 12V power in (from a power brick), and converts some of it to 5V & 3.3V for the PC, and passes some unconverted as 12V.  The PicoPSU's conversion from 12V to 5V/3.3V is very efficient, 94%+.  But some people are saying that the power brick itself, which converts the household AC to 12VDC, is only about 70% efficient.

 

It's true that while the PicoPSU itself is super efficient, the AC/DC power brick is less so.  That said, you can find power bricks that are much more efficient than standard ATX PSU's.  The power brick that is sold on mini-box's website in conjunction with a PicoPSU is listed as >87% efficient. 

 

The other part of the equation is when you use a power brick, it has a much lower max power draw than most ATX PSU's, so you're running the brick at a higher % load which gives better efficiency.  Basically running the power brick at say 30% load, instead of an ATX PSU at <10% load.  And as we all know, ATX PSU's running at less than 20% load are horribly inefficient.

 

Most tests I have seen of PicoPSU/power brick combos result in >=80% efficiency, even at low loadings (where most ATX PSU's get around 70%), and as high as 87% efficiency at moderate loadings.  Of course, if you can find a low enough wattage ATX PSU to keep you at >=20% load at idle (while still having enough headroom for the peaks associated with spinup) you'll get 80%+ efficiency that way, easier and cheaper.

 

Yes, the Pico efficiency is tricky to completely determine.  I've sent them a note to see if they can provide a more detailed efficiency curve for the 12v power bricks they offer => the 192W unit that's listed at 87% "average" sounds good ... but of course the real question is what's the efficiency drop to at 10% load ... which is about what I'd be drawing when idle.    I also asked them if the peak load can be sustained for 3-5 seconds to support 6 drives during spinup.

 

Note also that the ~ 95% numbers for the other loads are artificial => yes, they really do achieve that level of conversion efficiency; but the power they're converting is coming in at ~87% , so 95% is really 95% x 87% = 82.6% 

 

But it's still intriguing.  If the efficiency is > 80% for loads as low as 20W it would indeed be a nearly perfect unit for small, low-power systems like our PC-Q25B builds ... and the space it would free in the case is amazing !!  (even if you Velcro the power brick in the case, it'd still be virtually empty).

 

Yes, the Pico efficiency is tricky to completely determine.  I've sent them a note to see if they can provide a more detailed efficiency curve for the 12v power bricks they offer => the 192W unit that's listed at 87% "average" sounds good ... but of course the real question is what's the efficiency drop to at 10% load ... which is about what I'd be drawing when idle.    I also asked them if the peak load can be sustained for 3-5 seconds to support 6 drives during spinup.

 

Note also that the ~ 95% numbers for the other loads are artificial => yes, they really do achieve that level of conversion efficiency; but the power they're converting is coming in at ~87% , so 95% is really 95% x 87% = 82.6% 

 

But it's still intriguing.  If the efficiency is > 80% for loads as low as 20W it would indeed be a nearly perfect unit for small, low-power systems like our PC-Q25B builds ... and the space it would free in the case is amazing !!  (even if you Velcro the power brick in the case, it'd still be virtually empty).

 

Cool, can't wait to see what they say, please post their response.  As long as it can sustain spinup, it would be a perfect solution for my server and would free up soooo much room in the case.  Even at 80% efficiency at idle that would be a roughly 10% increase in efficiency over what I'm getting at the moment.

 

I know AHCI has provisions for staggered spinup, but I believe it's an optional feature and not required.  I don't recall seeing any settings in the P8H77-I BIOS, but I'll have to take another look.  The Seagate spec sheets state 2.0A for spinup, so roughly 24W/drive.  With 5 drives all spinning up at once that would be 120W just for the drives alone.  It'll certainly be close.

... please post their response.

 

Of course 

Out of interested do you need that speed of the 2760A? When buying this card it has RAID functionality is not better to use that then Unraid?

 

That's a good question, and still one I'm trying to answer.

 

I'm not going for ultra high performance (of which the 2760A is certainly capable, no doubt).  If I was using this for non-linear video editing or a high demand database, yes, I would want ultra high performance, and I would use the on-board RAID capability.  Doing so would require all drives to spin up for every access, though, so this would also be a high power consumption server.  Typical RAID redundancy only allows for the loss of 1 or 2 drives, any additional drive losses and you lose your entire array, so you better have some good backups!

 

With unRAID, only the drive that contains the data you're reading spins up, plus parity if you're writing, so it is very energy efficient (except for parity checks/rebuilds).  unRAID can handle a single drive failure with no data loss, but if you have additional drive failures you only lose the data on those drives.  Unlike RAID, which stripes your data across all drives, unRAID keeps your data whole and intact on a single drive.  So if I lost 2 drives on a 24 drive server, I've only lost 2 drives of data, and not all 24.  This approach also minimizes capacity lost to parity.  All of these features match my requirements perfectly.

 

I would happily pay less for a 2760A without the RAID features.

 

So, what is my performance goal, then?  Well, I want an unRAID server that performs unthrottled.  One of the most crucial tasks with an unRAID server is regular Parity Checks.  With less than optimal throughput, Parity Check time increases. Ditto for data rebuilds in case of a failure or drive upgrade.  An unRAID server will never, never be faster than a single drive or the network connection.  But a less than optimal hardware build can lead to an unRAID server that performs much slower than a single drive.  These problems only get more common the larger you scale up the unRAID build.  At 24 drives, I'm trying to be careful not to throttle the performance with unwise hardware choices.

 

In my own unscientific tests (more anecdotal than anything), I've noticed parity check processing time increase when I've provided less than 250MB/s per drive.  I can't provide more than 250MB/s with my current build, so I don't know if I'm at peak performance or not, but my hunch is that I'm not.  That's why I picked the 2760A, as it will give me 333MB/s per drive of bandwidth.  Is it really needed?  Is it enough?  I have no idea...

 

A good learning example is that garycase's parity check takes ~8 hours on a 3TB parity - he has all drives connected directly to the MB, so his configuration is optimal.  I too have a 3TB parity (same Red 3TB drives), and my parity checks are closer to 12 hours.  My drives are connected through 4 Adaptec 1430SA PCIe 1.0 x4 adapters, so each drive is bandwidth limited to 250MB/s, plus 5 more drives on the MB which are not bandwidth throttled.

 

While I can't say for sure exactly where my performance is hindered (CPU, memory, bus bandwidth, too many components, a high drive count, slow HD's, inefficient drivers, etc.), I'm certainly giving up some performance somewhere. 

 

Parity Checks and rebuilds also take longer the larger your parity drive.  I've watched my parity checks increase from a few hours with 1TB drives to 12+ hours with 3TB drives as my system has grown.  What happens with 6TB drives?  24 hour parity checks?  Yikes! Where does it end...

 

Another consideration is heat.  Drives are going to heat up when spun up - but my drives are spun up 50% longer than garycase's on a parity check.  Assuming similar cooling, who's drives do you think will fail first?

 

Drives are also continuing to get faster as platter densities increase.  What may not be a bottleneck with today's 3TB drives may be a bottleneck with tomorrow's higher performing 6TB drives (oh, did I say tomorrow... I meant 2016's).

 

So my performance goal is simply to have a great performing unRAID server, both today and 5 years from now, while minimizing drive wear and tear and energy consumption.

 

Sorry for my long winded-ness.  I probably didn't even answer your question...

A good learning example is that garycase's parity check takes ~8 hours on a 3TB parity - he has all drives connected directly to the MB, so his configuration is optimal.  I too have a 3TB parity (same Red 3TB drives), and my parity checks are closer to 12 hours.  My drives are connected through 4 Adaptec 1430SA PCIe 1.0 x4 adapters, so each drive is bandwidth limited to 250MB/s, plus 5 more drives on the MB which are not bandwidth throttled.

 

While I can't say for sure exactly where my performance is hindered (CPU, memory, bus bandwidth, too many components, a high drive count, slow HD's, inefficient drivers, etc.), I'm certainly giving up some performance somewhere. 

 

Agree with your comments r.e. native RAID on the 2760A vs. UnRAID.  Clearly for max performance you'd use the card's RAID ... but in addition to providing higher performance, it would also increase power consumption (a fair amount) and increase risk (much more impact with a multi-drive failure).

 

r.e. your quoted comments on parity check speeds => While the 1430SA's MAY be having some impact on your parity check speeds;  I suspect the MAJOR contributing factor is almost certainly the mix of drives.  As I'm sure you know, all modern drives use zoned sectoring -- the result of which is MUCH better performance on the outer cylinders than the inner ones.  Usually they'll retain very good performance for ~ 50% of the capacity;  then begin to throttle notably; and then drop significantly for ~ the last 20% or so.    With a mix of drives (1.5TB, 2TB, and 3TB according to the list in your sig); youll encounter this throttling 3 times during a parity check ... as the 1.5TB drives slow down;  then again for the 2TB; and finally for the 3TB.  If you watch your parity check speed, you'll clearly see this -- it will start off fairly high;  then begin slowing appreciably around 1.2TB "in" to the process ... getting slower and slower until it crosses the 1.5TB point;  then start slowing again at around 1.6TB, and getting slower until it crosses the 2TB point;  and then pick back up until the 3TB drives approach their innermost cylinders.

 

On my system, ALL of the drives are 3TB Reds, so this slowdown is uniform and only happens once.

 

In addition, the older drives likely have lower platter densities (the WD Reds are 1TB/platter -- your other drives are probably 500MB, 667MB, or 750MB/platter => so the data transfer rates are also slower independent of which cylinders you're reading.

 

 

A good learning example is that garycase's parity check takes ~8 hours on a 3TB parity - he has all drives connected directly to the MB, so his configuration is optimal.  I too have a 3TB parity (same Red 3TB drives), and my parity checks are closer to 12 hours.  My drives are connected through 4 Adaptec 1430SA PCIe 1.0 x4 adapters, so each drive is bandwidth limited to 250MB/s, plus 5 more drives on the MB which are not bandwidth throttled.

 

While I can't say for sure exactly where my performance is hindered (CPU, memory, bus bandwidth, too many components, a high drive count, slow HD's, inefficient drivers, etc.), I'm certainly giving up some performance somewhere. 

 

Agree with your comments r.e. native RAID on the 2760A vs. UnRAID.  Clearly for max performance you'd use the card's RAID ... but in addition to providing higher performance, it would also increase power consumption (a fair amount) and increase risk (much more impact with a multi-drive failure).

 

r.e. your quoted comments on parity check speeds => While the 1430SA's MAY be having some impact on your parity check speeds;  I suspect the MAJOR contributing factor is almost certainly the mix of drives.  As I'm sure you know, all modern drives use zoned sectoring -- the result of which is MUCH better performance on the outer cylinders than the inner ones.  Usually they'll retain very good performance for ~ 50% of the capacity;  then begin to throttle notably; and then drop significantly for ~ the last 20% or so.    With a mix of drives (1.5TB, 2TB, and 3TB according to the list in your sig); youll encounter this throttling 3 times during a parity check ... as the 1.5TB drives slow down;  then again for the 2TB; and finally for the 3TB.  If you watch your parity check speed, you'll clearly see this -- it will start off fairly high;  then begin slowing appreciably around 1.2TB "in" to the process ... getting slower and slower until it crosses the 1.5TB point;  then start slowing again at around 1.6TB, and getting slower until it crosses the 2TB point;  and then pick back up until the 3TB drives approach their innermost cylinders.

 

On my system, ALL of the drives are 3TB Reds, so this slowdown is uniform and only happens once.

 

In addition, the older drives likely have lower platter densities (the WD Reds are 1TB/platter -- your other drives are probably 500MB, 667MB, or 750MB/platter => so the data transfer rates are also slower independent of which cylinders you're reading.

 

Very much the case.  My drives are a mix of 2TB and 3TB Seagates.  I experience a slowdown around 1.6TB, but once I cross the 2.0TB threshold things speed up again until roughly the 2.5TB point, where it slows again until it finishes.  Of course these are not sudden slowdowns, but gradual.  As stated by garycase, the last 20% or so of the drive is the slowest part by far.

 

That said, those Seagate 1TB/platter drives are crazy fast (even in the last 20% of the drive where they slow the most) and my array finishes a parity check in right at 6 hours, average of 138MB/s.  2x 3TB and 2x 2TB (all 1TB/platter drives).

 

Last parity check:

Last checked on Tue Apr 30 06:02:06 2013 PDT (eighteen days ago), finding 0 errors.
> Duration: 6 hours, 2 minutes, 6 seconds. Average speed: 138.1 MB/sec

r.e. your quoted comments on parity check speeds => While the 1430SA's MAY be having some impact on your parity check speeds;  I suspect the MAJOR contributing factor is almost certainly the mix of drives.  As I'm sure you know, all modern drives use zoned sectoring -- the result of which is MUCH better performance on the outer cylinders than the inner ones.  Usually they'll retain very good performance for ~ 50% of the capacity;  then begin to throttle notably; and then drop significantly for ~ the last 20% or so.    With a mix of drives (1.5TB, 2TB, and 3TB according to the list in your sig); youll encounter this throttling 3 times during a parity check ... as the 1.5TB drives slow down;  then again for the 2TB; and finally for the 3TB.  If you watch your parity check speed, you'll clearly see this -- it will start off fairly high;  then begin slowing appreciably around 1.2TB "in" to the process ... getting slower and slower until it crosses the 1.5TB point;  then start slowing again at around 1.6TB, and getting slower until it crosses the 2TB point;  and then pick back up until the 3TB drives approach their innermost cylinders.

 

On my system, ALL of the drives are 3TB Reds, so this slowdown is uniform and only happens once.

 

You're spot on, and I'm well familiar with that behavior.  With my current mix, it is impossible for me to verify how much of the slow parity check is drive mix related, and how much is controller or other factor related.

 

I'm curious though, what average parity speed do you see reported at, say, 70%?  At the 70% mark, my parity check is fully on 3TB Red drives.  I'll have to check the reported speed again myself.

I'll run a parity check this evening and note what the speed is at some point after 70%.

 

FYI I just started a parity check -- should be 2/3rds of the way in 5 hrs or so ... I'll post the speed as close to 68% as I can (I'll check it fairly regularly).

 

Depending on what time zone you're in, you may want to start one so you'll be up around the time you expect it to hit that same %