HPE Storage Users Group

Got an odd one here gents, we appear to be experiencing some performance issues with our new/'empty' StoreServ 7400.

We are using 900GB 10K disks in the array, rated at +/- 140 IOPs each. We have 68 drives which equates to 9520 total RAW IOPs.

I’ve done some calculations against the functional IOPs based on the RAID Set for the 3PAR CPG (Raid 5 (3+1)). I’ve assumed average read/write percentage at 70% read, 30% write (these tend to be our workloads).

Raw IOPS = Disk Speed IOPs x Number of disks (= 9520)
Functional IOPs=((RawIOPs*Write %)/RAID Write Penalty)+(RawIOPs*Read %)
F= 9520 x 50% / 4 + 9520 x 50% (assuming read/write ratio is 50%)
F= 5950 IOPs

Now, due to the way the 3PAR creates its CPG’s and parity type (i.e. RAID SET) will depend on how many disks are used. I lose 4 disks between for the node pairs so that leaves 64 usable.


StoreServ 7400 –

Raw IOPS = Disk Speed IOPS x Number of disks (=8960)
Functional IOPS=((RawIOPS*Write %)/RAID Write Penalty)+(RawIOPS*Read %)
F= 8960 x 30% / 4 + 8960 x 70% (assuming read/write ratio is 30:70)
F= 6944 IOPs

For comparison I've included a Dell iSCSI array :

Dell Equallogic 6000XV -
Raw IOPS = Disk Speed IOPS x Number of disks (=2750)
Functional IOPS=((RawIOPS*Write %)/RAID Write Penalty)+(RawIOPS*Read %)
F= 2750 x 30% / 4 + 2750 x 70% (assuming read/write ratio is 30:70)
F= 2131 IOPs


Now, doing some tests from the same virtual machine across the 3PAR and iSCSI array’s.

Test 1a:
3PAR.CPG.RAID5(3+1)
MAX theoretical IOPs = 6944 IOPS
I/O size 4KB
100% Sequential
Read/Write = 70/30
IOPs achieved = 1124.25

Test 2a:
3PAR.CPG.RAID5(3+1)
MAX theoretical IOPs = 6944 IOPS
I/O size 4KB
50% Sequential : 50% Random
Read/Write = 70/30
IOPs achieved = 665.50

-----------------------------
Test 1b:
DellEQL.RAID50
MAX theoretical IOPs = 2750 IOPS
I/O size 4KB
100% Sequential
Read/Write = 70/30
Runtime 5 minutes
IOPs achieved = 2110.40

Test 2b:
DellEQL.RAID50
MAX theoretical IOPs = 2750 IOPS
I/O size 4KB
50% Sequential : 50% Random
Read/Write = 70/30
Runtime 5 minutes
IOPs achieved = 278.24

In test 2b the Equallogic performs poorer than the 3PAR as expected, but in Test 1a - the 3PAR should get close to its total functional IOPs (as observed in the Equallogic).

I've raised a support request with HP, but in the mean time I wonder if anyone had seen similar issues.

What test tool are you using. I've also got an empty 7400 4-node with 48 900 GB 10K SAS drives
iSCSI SAN, RAID 1 instead of RAID 5

I'd be interested in repeating the test to see is it a generic 3PAR issue or a config issue specific to your environment.

Hi, the tool used is IOmeter (running inside a vmware virtual machine).

Loading several worker processes (ie 10) yielded a significant improvement.

# Worker processes = 10
I/O Transfer size = 16KB
100 % Sequential distribution
Read/Write percentage = 70/30
Runtime 5 minutes

IOPs obtained = 13175