Hello together.
My Name is Ralph, i am 24 years old and i am coming from Kerpen near Cologne, Germany so sorry in advance for my language skills Image may be NSFW.
Clik here to view.
We currently have kind of a performance problem on a customers new environment.
The environment contains the following equipment:
3x Dell PowerEdge R620, each with 2x 8core Xeon, 64GB RAM, 2x 4 port Intel NICs and ESXi 5.0.0 U2.
2x Dell PowerConnect 6224 Switches with Dell RPS-600 PowerSupply.
2x Dell PowerVault MD3220i, each with Dual 4 Port Gigabit Controller and 16x 300GB SAS 15k HDDs.
2x Dell PowerVault MD1200 Disc Enclosure for the MD3220i´s, each with 12x 3TB NearlineSAS 15k HDDs, connected to MD3220i via SAS.
The MD3220i´s and MD1200´s have both identical RAID configurations:
MD3220i: 10x 300GB @ RAID10 for Production environment, 4x 300GB @ RAID10 for Testing environment, 2x HotSpare
MD1200: 11x 3TB @ RAID6 for Files, 1x HotSpare
The network configuration is the following:
VLAN 42 has IP Range XXX.YYY.42.0 /24
VLAN 43 has IP Range XXX.YYY.43.0 /24
Switch 1: "SAN-ports" 3-16 configured for iSCSI (Jumbo frames, storm control, etc.pp) with Tagged VLAN 42.
Switch 2: "SAN-ports" 3-16 configured for iSCSI (Jumbo frames, storm control, etc.pp) with Tagged VLAN 43.
Servers 1-3: two ports of each NIC (four ports in summary) configured for iSCSI (Jumbo Frames) - two with VLAN 42, two with VLAN 43.
MD3220i´s: two ports of each controller configured with Jumbo Frames and VLAN 42, two with Jumbo Frames and VLAN 43.
Everything cabled for redundancy, each Server has 2 GBit to Switch 1 and 2 GBit to Switch 2 (even with a NIC-fail there will still be 1 GBit to each switch), so as the MD3220i´s.
Theoratically there should be a bandwith of 4 GBit from e.g. Server 1 to MD3220i.
Alle Volumes are set up und initialized.
A Windows Server 2008 R2 VM is installed on the 15k Production Pool (10x 300GB @ RAID10) on first MD3220i and has a 18TB LUN (RAID6) attached via RDM from first MD1200.
A Debian VM is installed on the 15k Production Pool (10x 300GB @ RAID10) on second MD3220i and has a 18TB LUN (RAID6) attached via RDM from second MD1200.
Now the performance issue(s):
When copying a huge test file in Windows and also Debian VM (20GB) from 18TB RDM to HDD in Production Pool (vmdk) there is only near 200 MB/s, often much below this (average seems to be 110-120MB/s)... Once i reached an average of 350MB/s with robocopy on Windows VM (this speed is fine and my "target"!)
When copying a huge test file in Windows and also Debian VM (20GB) from HDD in Production Pool (vmdk) to 18TB RDM there is a maximum rate of 85 to 90 MB/s...
Same performance when writing/reading from one storage to another.
When i check esxtop on the VMs host while copying the file i can see that RoundRobin is working because the load is "balanced" on all 4 NICs.
But they always don´t reach more than 1 GBit/s in summary.
The target is to reach nearly 4 GBit/s everywhere (each combinations, from 15k to 7,2k HDDs, from 7,2k to 18k HDDs, from storage1 to storage2, etc.pp.).
I think it is easier in RAID10 than in RAID6, but IMO there are enough spindles to manage that.
Will that be possible?
Is there a bottleneck anywhere?
Looking forward hearing from you!
Kind regards
Ralph