Unsolved
This post is more than 5 years old
8 Posts
0
1346
February 21st, 2017 14:00
MD3620i slow performance with 2012 R2 Hyper-V 2-node Cluster
Hi,
We have the following set up:
1x DR4100 (2x 10GbE Domain Team) - provides virtual tape library to Microsoft DPM 2016 (virtual guest)
2x R730:
2x 300GB system disk in RAID-1
4x 10GbE QLogic BCM57810 (2x dual port adapters)
4x 1Gb Broadcom NetXtreme (onboard, unused)
QLogic Driver 7.13.54.0 25/03/2016, Firmware 7.12/7.14 (multiple versions listed in QLogic Control Suite)
Card 1 Port 1 - iSCSI VLAN
Card 1 Port 2 - Domain Team
Card 2 Port 1 - iSCSI VLAN
Card 2 Port 2 - Domain Team
Domain Team is used for, Guest Communication, Cluster Communication, Host Management.
MTU is set to 1500 for all adapters, flow control is off.
These 2 hosts share CSV's on a MD3620i and MD1200
Md3620i - 22x 300GB SAS 15K, 1x RAID-10, 2 LUNs, Dual controllers (4x 10GbE, 2x per controller)
MD1200 - 20x 1TB SATA 7200,2x RAID-10 evenly split, 1 LUN per RAID array, connected to MD3620i with 2x 6Gb SAS
Arrays are 64k segements, formatted to 64k ntfs
I'm finding performance is slow when doing internal copying from LUN to LUN, even within the same LUN, or from MD1200 to MD3620i.
If I copy from Host to MD I get 1.2GB/sec, but moving data around on the LUNs gets me about 80-150MB/sec - surely I should be getting more than this with RAID-10?
When running DPM, backup speeds tend to max out about 500Mb/sec during a host-level VM backup.
I Would expect speeds of around 1GB/sec when accessing the MD3620i CSVs and 300MB/sec from the MD1200 CSVs due to the 6Gb SAS limitation.
When migrating VM storage from one LUN to another, using the failover cluster manager, seems to take forever also.
Any ideas what could be causing my slow speeds? Am I expecting too much from the MD?
Thanks
F.James
8 Posts
0
March 5th, 2017 06:00
Still trying to resolve this.
So far I have done the following:
Upgraded MD3620i to 08.20.21.61, MD1220 ESM to 1.06, HDD from AS0B to AS0C
Upgraded MDSM to 11.25.0G0.0026 using Resource DVD DELL_MDSS_Consolidated_RDVD_6_5_0_1.iso
Enabled jumbo frames on switch, iSCSI VLAN
Enabled Jumbo 9000 MTU on the MD3620i controller ports
Enabled flow control on the switch and on the QLogic BCM57810 NDIS iSCSI ports.
Enabled Jumbo frames on the same NIC ports, 9014 bytes from the adapter properties - done on both vHosts
1. Why don't the MDSM versions match? Did I use the right installer?
When running the main installer from the root of the disk it only seems to install the disk configuration client, rather than the full MDSM (even though full install is selected). I had to browse to windows\mdsm and run SMIA-Winx64.exe for a full install.
Even after this there are some version mismatches.
MDSM (Enterprise Management) shows version 11.25.0G0.0026
If I manage the array from this point, it loads the MDSM (Array Management) - Summary window shows Management Software Version 11.20.0.G0.6.0020 and Help>About confirms this.
2. Why cant I ping the controller ports with the higher MTU size after enabling Jumbo frames at the switch/NIC/MD?
I can ping -f -l 8000 between the two hosts and the different cards, but when pinging the MD3620i controller ports I get request timed out. Running a standard ping gets a response from the MD3620i