This post is more than 5 years old
5 Posts
0
973
iSCSI Multipath problem with CX4 and Windows 2008
I am having a problem with PowerPath 5.3 SP1 and a Win 2008 server. After installing powerpath I am only get 29 IOPS per used path with occasional spikes to up to 130. Only two path are used out of the four which is to be expected. If I set all but one path to be in standby the IOPS on the only active channel goes up and maintains at least 680 IOPS. This does not seem to be related to a specific channel or controller because ever path can consistently sustain 680 IOPS if is the only channel used.
The server has to NIC's (BC5716) in it, each attaching to a different iscsi network. The networks are 10.70.160.0/24 on one NIC and 10.70.140.0/24 on the other.
The CX4 has 4 iscsi ports total, 2 on each controller. SPA2 and SPB2 are connected to 10.70.140.0/24 and SPA3 and SPB3 are connected to 10.70.160.0/24.
I have the made all the changes in the registry that are recommended by EMC and have the latest firmware and drivers for the NIC's
Any suggestions?
kelleg
4.5K Posts
0
May 26th, 2010 13:00
Under certain conditions a local disk can performance better than an array LUN (overhead for network, remote cache vs on-board cache, disk caching locally, etc.). Using 15K FC disks, each disk can handle a sustain workload of about 180 IOPS or 12MB/s for larger IO's (larger than 32KB). The more sequential the better the performance - Reads vs. Writes, random vs sequential, small block vs large block.
When you create a Raid 5 raid group using 5 disks, a 4+1, a single LUN in this raid group can handle 5 * 180 IOPS = 900 IOPS or 12MB/s * 5 = 60MB/s.
For more detailed information - look in the Documents section on the forum and find the Best Practices guide for the version of flare on your array. There is a section in there about determining the number of disks required to handle a workload.
glen
kelleg
4.5K Posts
0
May 25th, 2010 12:00
The server has to NIC's (BC5716) in it, each attaching to a different iscsi network. The networks are 10.70.160.0/24 on one NIC and 10.70.140.0/24 on the other.
The CX4 has 4 iscsi ports total, 2 on each controller. SPA2 and SPB2 are connected to 10.70.140.0/24 and SPA3 and SPB3 are connected to 10.70.160.0/24.
NIC1 10.70.160.x <---> SPA3 SPB3
NIC2 10.70.140.x <---> SPA2 SPB2
Are you using licensed PowerPath? If so, you should see IO down both NIC's for a single LUN.
Have you looked at Knowledgebase article emc150702 - this is for the TCPDelayed Ack issue in Windows.
glen
DanJost
190 Posts
0
May 26th, 2010 06:00
Have you read the tech note on iSCSI and data traffic? It has some pretty specific guidelines in there...It mentions some possible configurations where the traffic would be directed out of the management interface which would probably be slow if your SPs are connected at 100Mbit
The document is under Technical Notes and Troubleshooting->Managing Clariion iSCSI Data and Control Traffic
Dan
reagleston
5 Posts
0
May 26th, 2010 10:00
I will look at the article. I am fairly certain though that the traffic is going out on the correct NICs. The management and LAN interface are on seperate interfaces and on different VLANS. I verified with tracert that the IP traffic to the Clariion's controllers was not being routed. All of the registry changes have also been made, ie TcpAckFrequency, etc. I have also verified that jumbo frames are being used.
I updated the Broadcom drivers and now have both NIC performing at 300 IOPS ( or 70MB/sec) while performing 512K reads. This seems still seems slow. The server has a single 7.2K 1TB SATA drive that benchmarks at 90MB/Sec on the same test. The LUN that is attached through iSCSI is RAID 5 set of 5 15K FC drives. Shouldn't I see better performance even with iscsi because the of the 5 FC Drives in the LUN over a single SATA drive that is direct attached?