Server reshuffle

[pras1011]

Still flying but I have noticed that disk5 has a lot less reads.

 

Is this ok?

[garycase]

Yes, that's fine.  The read counts can vary all over the place ... it's "magic" 

 

Actually it has something to do with size of the actual read requests, which can vary based on available buffer space, etc.    In earlier versions of UnRAID (and with IDE controllers), the read counts were uniform; but a few years ago they started being all over the map.  I've asked Tom about this, and never have really understood exactly why they vary so much -- but it IS normal.

 

[pras1011]

8tb parity build in 54875 seconds in 146MB/s. Not bad! Now I am doing a parity check.

[garycase]

8tb parity build in 54875 seconds in 146MB/s. Not bad! Now I am doing a parity check.

 

"Not bad" ??    Quite an understatement !! 

 

[pras1011]

Parity build was rapid but the parity check is slow normal speed though.

[garycase]

Parity build was rapid but the parity check is slow normal speed though.

 

How "slow" ??    I'd still think it would be pretty fast compared to systems with lower areal density drives.

 

[pras1011]

About 120

[garycase]

About 120

 

That's not, of course, "slow" -- just slower than the sync was 

 

[pras1011]

Haha yes. So glad the transfer is over! Now the transfer from one server case to another will start soon. Shouldn't be too difficult.

[pkn]

Parity build was rapid but the parity check is slow normal speed though.

It's normal. Because in  parity sync there is only one active process mdrecoveryd, and in parity check there are two: mdrecoveryd and unraidd, competing for the CPU cycles.

[pras1011]

Thanks.

 

Talking of cpus, I think the g645 is a bit crap. Whats a good cpu to replace this with?

[HellDiverUK]

Thanks.

 

Talking of cpus, I think the g645 is a bit crap. Whats a good cpu to replace this with?

 

He he, I use a G620 in my remote access machine!  Speed demon there.   Though it's passively cooled and sips power.

 

With your board, you're looking at the likes of an i3-3220 or even an i5.  I have run UnRAID on my i5-3570S which ran very well - 3.1GHz Quad core with turbo to 3.8GHz and 65W TDP.

 

I'm actually thinking about going back to that chip on my venerable Asus B75-M board, it's a truck, totally unbreakable.

 

 

[pras1011]

I was thinking of the i3 but I bought myself a cheap g2030.

[HellDiverUK]

I doubt you'll notice any difference between the two.  100MHz difference - the G645 is dual 2.9GHz Sandy Bridge, the G2030 is dual 3GHz Ivy Bridge.

 

http://ark.intel.com/compare/69116,74749

 

Only possible improvement is the DDR3-1333 compared to DDR3-1066 memory, and even that's nothing to write home about.

[pras1011]

Well the parity check finished in 18.5 at an average of 120MB/s.

[pras1011]

I know the g2030 is not that much better than the g645 but I got it dirt cheap. It also uses 10 watts less than the g645 as well as the ddr3-1333.

 

I have has a strange problem: when I drag and drop a bluray file to copy to the server it timeouts. Basically the server does a lot of reading on the target disc. However if I use makemkv to transfer it, it starts straight away. WTF?

[garycase]

... I know the g2030 is not that much better than the g645

 

Agree ... the g2030 scores 2924 on PassMark;  the g645 2604.  So the G2030 is only a 12% bump in performance ... not enough that you're likely to notice it.

 

A Core i3-4160 scores 5049 ... almost twice the performance of a g645 => THAT you WOULD have noticed 

... but clearly we all make our own price/performance choices -- and I suspect the G2030 will work just fine for you.

 

 

[pras1011]

The 4160 is 1150 and therefore won't work on my mobo. However I have been thinking about getting a new 1150 mobo with 8 plus sata ports. 7 x 8tb is 56tb and hopefully should be more than enough for me so that I can get rid on the sas2lp. Less components the better I would say.

 

But 6 of the ports will run on the Intel controller and the other 2 will be asmedia or something similar which I think won't be great.

[garycase]

I forgot you had an 1155 board => in that case an i3-3250 would have been a good choice if you wanted significantly more "horsepower" without spending the $ for an i5 or i7 class CPU.    It scores 4517 on PassMark, not quite as good as a Haswell, but without the cost of a new motherboard.

 

http://www.superbiiz.com/detail.php?p=I3-3250BOX&c=fr&pid=5be0eb858cedecdf2593afa88e982b476ed215842bbb0f208222f0e7096a9467&gclid=COWu75ee6sUCFQioaQodRBsAgw

 

However ... a nice new Haswell setup would not only give you more power, but would also be more power efficient and run cooler.     

 

And I agree that an 8-port motherboard with 8TB drives can provide enough storage that most folks won't need any add-on controllers.    I don't think the 2nd SATA controller would matter as long as it's a controller that's supported by UnRAID ... as far as I know most of the secondary controllers used on typical motherboard ARE supported.    That also allows using a significantly smaller case than you'd need for equivalent storage capacity with smaller drives.

 

[pras1011]

Would a on board controller be faster than a external one?

[garycase]

Would a on board controller be faster than a external one?

 

I presume by "external one" you mean an add-on board.    The answer is "it depends" ==> as long as the add-in card is in a slot that has enough bandwidth to support all attached drives at full speed it makes no difference.  For example, the popular cards that use PCIe x4 or x8 interfaces can easily support 8 drives at full bandwidth.

 

But if you use an older PCI card with 4 ports  on it, then the card will be a bottleneck that slows down the system whenever all of the drives are being used at once [e.g. a parity check or rebuild].

 

In addition, when ANY port ... whether onboard or on an attached card ... is used with a port multiplier,  all the attached drives on the multiplier share the bandwidth of one SATA port, so this can also slow things down a bit.

 

[HellDiverUK]

As garycase says, depends on the controller.  Cheaper Marvell cards tend to be a bit pants.  They're the Realtek of the SATA controller world.

 

[pras1011]

I have the gene v and a sas2lp. I have plugged it into the pcie slots. 4 of the sata ports are run by Intel and the other two are run by Marvell. I though a direct connection of a hdd to the mobo will always be faster than an indirect one.

[garycase]

Just depends on the bandwidth => as long as you have enough bandwidth to support full speed transfers to/from the disk drives, it makes no difference which SATA port you're using.

 

 

[pras1011]

So does not matter if it's direct or indirect. It must save a bit of power not using the sas2lp?

 

Is the speed of the z77/z97 chipset ultimately determined by the power of the cpu? If yes then better not get a pentium. Or is the chipset platform hub controller determining the speed?