Thursday, March 31, 2011

Configuring Dell PowerConnect 6224 Switches for iSCSI traffic to a Compellent

This is the second time I have set up a Compellent, and this time I figured I would go all the way down the rabbit's hole and delve into switch optimization for iSCSI traffic. My last project was with Cisco switches, this time, it's with Dell. A pair of PowerConnect 6224 switches to be precise... main reason being, Dell purchased Compellent not long ago, so like-products, yadda yadda. Alas, I digress.

In previous talks with Compellent, I was give the '10 commandments' of switch configuration from at least two different representatives upon having iSCSI issues. These commandments roughly consisted of:

• Gigabit Full Duplex connectivity between Storage Center and all local Servers
• Auto-Negotiate for all switches that will correctly negotiate at Gigabit Full Duplex
• Gigabit Full Duplex hard set for all iSCSI ports, for both Storage Center and Servers for switches that do not correctly negotiate
• Bi-Directional Flow Control enabled for all Switch Ports that servers or controllers are using for iSCSI traffic.
• Bi-Directional Flow Control enabled for all Server Ports used for iSCSI (Storage Center and QLogic HBA's automatically enable it).
• Bi-Directional Flow Control enabled for all ports that handle iSCSI traffic. This includes all devices in between two sites that are used for replication.
• Separate VLAN for iSCSI.
• Two separate networks or VLANs for multipathed iSCSI.
• Two separate IP subnets for the seperate networks or VLANs in multipathed iSCSI.
• Unicast storm control disabled on every switch that handles iSCSI traffic.
• Multicast disabled at the switch level for any iSCSI VLANs.
o Multicast storm control enabled (if available) when multicast can not disabled.
• Broadcast disabled at the switch level for any iSCSI VLANs.
o Broadcast storm control enabled (if available) when broadcast can not disabled.
• Routing disabled between regular network and iSCSI VLANs.
• Do not use Spanning Tree (STP or RSTP) on ports which connect directly to end nodes (the server or Compellent controllers iSCSI ports.) If you must use it, enable the Cisco PortFast option on these ports so that they are configured as edge ports.
• Ensure that any switches used for iSCSI are of a non-blocking design.
• When deciding which switches to use, remember that you are running SCSI traffic over it. Be sure to use a quality managed enterprise class networking equipment. It is not recommended to use SBHO (small business/home office) class equipment outside of lab/test environments.
• Verify optimal MTU for replications. Default is 1500 but sometimes WAN circuits or VPNs can create additional overhead which can cause packet fragmentation. This fragmentation can suboptimal performance. The MTU is adjustable via the GUI in 5.x Storage Center Firmware.
For Jumbo Frame Support
• Some switches have limited buffer sizes and can only support Flow Control or Jumbo Frames, but not both at the same time. Compellent strongly recommends choosing Flow Control.
• All devices connected through iSCSI need to support 9k jumbo frames.
• All devices used to connect iSCSI devices need to support it.
o This means every switch, router, WAN Accelerator and any other network device that will handle iSCSI traffic needs to support 9k Jumbo Frames.
• If the customer is not 100% positive that every device in their iSCSI network supports 9k Jumbo Frames, then they should NOT turn on Jumbo Frames.
• QLogic 4010 series cards (Early Compellent iSCSI Cards) do not support Jumbo Frames.
o In the Storage Center GUI default screen, expand the tree in the following order Controllers->SN#(for the controller)->IO Cards->iSCSI->Highlight the port and the general tab should list the model number in the description.
• Because devices on both sides (server and SAN) need Jumbo Frames enabled, the change to enable to disable Jumbo Frames is recommended during a maintenance window. If servers have it enabled first, the Storage Center will not understand their packets. If Storage Center enables it first, servers will not understand its packets.

Okay, so it was more than 10. Anyway, all that roughly distills down to the following needs:

1) Set spanning-tree mode rstp and enable portfast on all ports, or disable spanning-tree all together.
2) Enable jumbo frame support on all iSCSI ports
3) Disable unicast storm-control on all iSCSI ports
4) Enable multicast storm-control on all iSCSI ports
5) Enable broadcast storm-control on all iSCSI ports
6) Enable flow control

For each switch, the commands are different... but for Dell PowerConnect 6224 in particular, these needs translated into the following commands (in order, considering I used ports 1-12 for iSCSI)

spanning-tree mode rstp
interface 1/g1-1/g12
spanning-tree portfast
mtu 9216
no storm-control unicast
storm-control multicast
storm-control broadcast

That was it, switch optimized, on we go!