better read performance in storage spaces

[mineblaster]

be playing around  with 5 ssd in a stripe configuration and was wounding if there is tweaks to get better read performance.  it seems to be lacking

 

done some test with ATTO at the defaults and queue depth of 32

 

the system config is 

5x micron p320h 700gb (3.2gb read 1.9gb write)

 64gb of ram 

2620v2 x2 xeons

 

 

any tips are welcome

 

 

 

[NunoLava1998]

impressive numbers aside;

1. those sequential numbers are unnecessary

2. enjoy your 4K speeds going to 0

[SkyHound0202]

Unless you really need that much throughput and don't care about redundancy, it's never wise to run RAID 0.

 

That said, if you still want to go faster, try updating the SSD firmware and balance the PCIe lane usage between processors. Add cooling for the controller on the drives if needed.

 

But, just like the other guy said, these numbers are just pointless numbers which may not have a real life implication if your specific workload cannot take advantage of it.

[leadeater]

38 minutes ago, mineblaster said:

be playing around  with 5 ssd in a stripe configuration and was wounding if there is tweaks to get better read performance.  it seems to be lacking

Set the storage pool to IsPowerProtected? Write performance can often just be a lie, system memory gets in there. You should really be testing with IOmeter and more than 15 minute run time.

[Electronics Wizardy]

what is pool config? File system? number of collumns?

[mineblaster]

On 5/4/2019 at 2:01 PM, Electronics Wizardy said:

what is pool config? File system? number of collumns?

its a stripped config (raid 0) 5 collumns and default ntfs

[Electronics Wizardy]

13 minutes ago, mineblaster said:

its a stripped config (raid 0) 5 collumns and default ntfs

try anouther benchmark like dsk speed or iometer.

 

This might be a cpu limit aswell, getting that much io can use a lot of cpu power.

[mineblaster]

On 5/1/2019 at 10:16 PM, leadeater said:

Set the storage pool to IsPowerProtected? Write performance can often just be a lie, system memory gets in there. You should really be testing with IOmeter and more than 15 minute run time.

i dont have a battery backup configred at the moment so i dont think its enabled unless its by default.

any good setting for setting iometer i havent really used it before i looked it up alittle

[mineblaster]

24 minutes ago, Electronics Wizardy said:

try anouther benchmark like dsk speed or iometer.

 

This might be a cpu limit aswell, getting that much io can use a lot of cpu power.

i have crystal disk

http://prntscr.com/nl5yzp

 

but its not much better

also been on the hunt for cpus but can decide on 2667v2 or the 2687w v2.

 

i haven seen cpu usage over 10% so far

 

is the all setting good for iometer?

[Electronics Wizardy]

29 minutes ago, mineblaster said:

i dont have a battery backup configred at the moment so i dont think its enabled unless its by default.

any good setting for setting iometer i havent really used it before i looked it up alittle

bbu shouldn't matter for software raid. But you can set storage spaces to think there is a bbu, just try that ,but pobably won't help much.

 

19 minutes ago, mineblaster said:

i have crystal disk

http://prntscr.com/nl5yzp

 

but its not much better

also been on the hunt for cpus but can decide on 2667v2 or the 2687w v2.

 

i haven seen cpu usage over 10% so far

 

is the all setting good for iometer?

What is the full system config? What slots are being used?

 

Are the single drives getting the expected speed?

 

What os?

[mineblaster]

2620v2 x2

8x8gb ddr3 @ 1600MHz quad channel

the p320h ssds take up all the slots and are connected @pci 2.0 8x

supermicro X9DR3-F

and a 10gbe nic

 

os is server 2016

and the drives get full speed on their own

[Electronics Wizardy]

Just now, mineblaster said:

2620v2 x2

8x8gb ddr3 @ 1600MHz quad channel

the p320h ssds take up all the slots and are connected @pci 2.0 8x

supermicro X9DR3-F 

and a 10gbe nic

 

os is server 2016

and the drives get full speed on their own

Try server 2019 if you can, few good improvemnts in storage spaces.

 

Do you get full speeds on linux with something like md?

 

Im guessing there may be an issue with numa, using qpi between some drives, the single threaded performance of the chips. Really no way to test without a newer platform.

 

 

[mineblaster]

16 minutes ago, Electronics Wizardy said:

Try server 2019 if you can, few good improvemnts in storage spaces.

 

Do you get full speeds on linux with something like md?

 

Im guessing there may be an issue with numa, using qpi between some drives, the single threaded performance of the chips. Really no way to test without a newer platform.

 

 

ill load up 2019 tomorrow to see if the smaller file reads and writes pick up better.

also qpi is at 7.2 gigatransfers so it most likely be the cap for larger sequential transfers

 

the newer platform is a dream for the time being and also pretty much never used linux