Hello all,
I have a misunderstanding with a HP MSA2312i(iSCSI) with Dual-Controller and 4 giga ethernet ports and one ESX 4.1 Enterprise Plus on HP DL360G7 with 8 ethernet giga ports.
I have 2 vSwitches with one pNIC each other. One pNIC Conected to Controller0-Port1 on vSwitch0 and vSwitch1 with one pNIC conected to Controller1-Port1. One LUN is exported to ESX and is seen on 2 way, so multipath is working.
The problem is I cannot get more than 1Gbps read/write even the LUN is presented to ESX through 2 pNIC. I tried RoundRobin but still limited to 1Gbps read/write. The 2312i support controller Load balancing. Do you have any ideas to make it to work with more than 1 ethernet port and get more throughput?
Any ideas are welcome.
Thanks,
D.
Some useful readingon this very interesting subject:
Hope this helps
Paul Grevink
If you find this information useful, please award points for "correct" or "helpful".
Message was edited by: PaulusG now with correct URL vSPhere
What are you using to measure your throughput?
If you found this at all helpful please award points by using the correct or helpful buttons! Thanks!
I am using iometer and second method is to copy one iso file from VM1 located on a RAID10 to VM2 located in RAID5 and vice versa. The arrays are located on the same MSA2312i storage. I am sure the transfer can be more than 110MB/s and the bottleneck is the network.
Also I red and implented this solution but without success. still a rate transfer of 110MB/s. (http://yourmacguy.wordpress.com/2009/11/09/vsphere-iscsi-multipathing/ )
on vmhba37(iscsi software initiator) i have vmk0 with a vmnic0 and vmk1 with vmnic1 and datastore path is Round Robin.
Thanks
D.
110MB/s is about right for a single 1G connection. It's therectical max is 125MB/s, which you will never get. When I typically look at storage throughput it's in IOPS not MBs. Have you figured out what your IOPS requirements are and seen if the test will accomidate it?
Regarding the multiple Ports:
Did you bind those nics to the iSCSI adapter?
Did you read the Best practice recommendations for HP?
With their EVA product they recommend changing the IOPS threshold from 1000 for each nic to 1 with round robin.
How is the SAN side of things configured?
If you found this at all helpful please award points by using the correct or helpful buttons! Thanks!
On this LUN will reside entire MSSQL databases(~200GB) from ERP software. I dont know the necessary IOPS.
1. I Already have binded 2 vmk nics to iScsi software initiator, changed to RoundRobin path of Datastore and set IO Operation limit to 1.
esxcli --server ESX02 nmp roundrobin getconfig --device naa.600c0ff00010d4ae546ed84c01000000
Byte Limit: 10485760
Device: naa.600c0ff00010d4ae546ed84c01000000
I/O Operation Limit: 1
Limit Type: Iops
Use Active Unoptimized Paths: false
# esxcli swiscsi nic list -d vmhba37
vmk0
pNic name: vmnic1
ipv4 address: 10.11.10.124
ipv4 net mask: 255.255.255.0
ipv6 addresses:
mac address: d4:85:64:5f:54:c2
mtu: 1500
toe: false
tso: true
tcp checksum: false
vlan: true
vlanId: 0
ports reserved: 63488~65536
link connected: true
ethernet speed: 1000
packets received: 17780441
packets sent: 2248554
NIC driver: bnx2
driver version: 2.0.7d-2vmw
firmware version: 5.2.2 NCSI 2.0.6
vmk1
pNic name: vmnic0
ipv4 address: 10.10.10.111
ipv4 net mask: 255.255.255.0
ipv6 addresses:
mac address: d4:85:64:5f:54:c0
mtu: 1500
toe: false
tso: true
tcp checksum: false
vlan: true
vlanId: 0
ports reserved: 63488~65536
link connected: true
ethernet speed: 1000
packets received: 11909248
packets sent: 17684191
NIC driver: bnx2
driver version: 2.0.7d-2vmw
firmware version: 5.2.2 NCSI 2.0.6
Sometimes I got a transfer rate around 170MBps but not always. IOmeter still shows 110MB/s
2. I have read the HP Best practices from here http://www.hpcollateral.com/Files/EcoCollateral_20090420_Whitepaper_MSA_Best_Practice__ipcz.pdf
3. the storage has 4 Ethernet ports: 2 ports per controller A(A0,A1) and 2 ports per Controller B(B0,B1). A0 and B0 are connected to the server directly with cables. there is no switch between them.
Hi,
While the MSA2000 is an active/active array it adheres to the ALUA specifications. Therefore one path is considered preferred, the one on the owning controller. If you look at the output from esxcli you will notice Use Active Unoptimized Paths: false which means that roundrobin will not use the unoptimized path.
If you want to get more throughput then you need to create additional preferred paths to the owning controller on the other ports, i.e. A1 or B1. This will give you two preferred paths to a controller.
Hope this makes sense.
Kind regards,
Glen