Administrator Guide

Network Infrastructure Performance Recommendations
Network performance is complex and depends on a number of components working with each other. You might be able to improve
network performance by following these general recommendations:
• Make sure that network components are recommended for an iSCSI SAN by Dell EqualLogic.
• Make sure the switches and interswitch links have sucient bandwidth for the iSCSI I/O. Contact EqualLogic customer support
for details on correct sizing and conguration of iSCSI SAN switch infrastructure. Pay careful attention to sizing interswitch links.
• Make sure all member network interface connections are Gigabit Ethernet and make sure driver settings are correct on the
servers. (SAN Headquarters generates an alert if it detects connections that are less than 1GB.)
• Make sure you connect and enable all the member Ethernet interfaces to maximize the available SAN bandwidth. If all interfaces
are enabled but bandwidth is still insucient, increasing the number of arrays in the storage pool might provide additional
throughput if the servers have not reached their maximum bandwidth.
• If only one interface is completely utilized on a member or on a server with multiple NICs, ensure that MPIO is properly
congured. If all of the server NICs exceed their capacity (to determine this, use host-based tools such as SAN Headquarters)
but the PS Series group has excess network capacity, add additional server NICs. Also, congure MPIO for those operating
systems that support MPIO.
• Lack of receive buers can also cause network performance problems. Low-end switches often have limited memory and suer
from performance issues related to insucient buers. Dell recommends a 1MB per port buer level in switches and, preferably,
dedicated buers over shared buers.
• Server performance can often be improved by increasing the number of buers allocated to the server NICs. Consult your
switch vendor or server NIC driver documentation to determine if you can increase the buers.
• Network bandwidth saturation is typically not a problem if it occurs during sequential operations and if application performance
remains acceptable. To resolve problems with network bandwidth, you might have to redistribute the workload over multiple
arrays. The PS Series architecture permits bandwidth to scale, either by enabling additional array interface ports or by adding
another array to the group and thus adding more controllers and network ports.
• Make sure you also follow the network performance guidelines in the Installation and Setup Guide. In particular:
– If possible, do not use Spanning Tree Protocol (STP) functionality on switch ports that connect end nodes.
– Enable ow control on switches and NICs — The most critical function that aects network performance is ow control,
which allows network devices to signal the next device that the data stream should be reduced to prevent dropped packets
and retransmissions.
– Disable unicast storm control on switches.
– Enable jumbo frames with care — In environments with small average I/O sizes, using jumbo frames has limited benets. In
general, support for jumbo frames is disabled by default on switches and server NICs. Enabling jumbo frames requires that
the switch use a VLAN other than the default VLAN (usually VLAN 1). PS Series arrays automatically negotiate using jumbo
frames when the iSCSI connection is established by the server.
Consult your switch vendor or server NIC driver documentation to determine if jumbo frames can be congured. Note that
some network devices run more slowly with jumbo frames enabled, do not properly support jumbo frames, or cannot support
them simultaneously with ow control. In these cases, jumbo frames should be disabled, or the switches should be upgraded
or replaced.
PS Series Group Performance Recommendations
To improve performance on your PS Series groups, consider the following group conguration recommendations for these areas:
network interfaces, storage pools, RAID policy, volumes, hardware and rmware, iSCSI connections, and MPIO.
Network Interfaces
Make sure all the network interfaces on the members are congured, functioning, and accessible. Use the Group Manager CLI ping
command to check accessibility.
NOTE: If you need more network bandwidth for iSCSI I/O, do not congure a dedicated management network.
Storage Pool Capacity
Low storage pool capacity is a problem that generates an alert in SAN Headquarters. If a pool has less than 5 percent of free space
(or less than 100GB per member, whichever is less), a PS Series group might not have sucient free space to eciently perform the
348
About Monitoring