Troubleshooting Performance and Responsiveness Issues on the Cluster

If you are experiencing performance or responsiveness issues on the transfer cluster, review the following configurations:

What is the instance size?

The recommended node instance size is m4.xlarge or larger. For maximum throughput of up to 3Gbps, use a c3.8xlarge.

What are the configured transfer target rate and vlink settings?

For a transfer of 3Gbps, make the configuration changes below. Select Edit Transfer Configuration from the cluster Actions drop-down menu.
Note: The <capacity> parameter is measured in kilobits per second (kbps).

Set the following in the <trunks> section:

<trunks>
  <trunk>
    <id>70</id>
    <capacity>
      <schedule format="ranges">3000000</schedule>
    </capacity>
    <on>true</on>
  </trunk>
  <trunk>
    <id>80</id>
    <capacity>
      <schedule format="ranges">3000000</schedule>
    </capacity>
    <on>true</on>
  </trunk>
</trunks>

Set the following in the <transfer> section:

<transfer>
  ...
  <in>
    <bandwidth>
      <aggregate>
        <trunk_id>70</trunk_id>
      </aggregate>
      <flow>
        <min_rate>
          <lock>true</lock>
          <cap>0</cap>
        </min_rate>
        <policy>
          <default>fair</default>
          <allowed>fair</allowed>
        </policy>
        <target_rate>
          <default>700000</default>
          <cap>3000000000</cap>
        </target_rate>
      </flow>
    </bandwidth>
  </in>
  <out>
    <bandwidth>
      <aggregate>
        <trunk_id>80</trunk_id>
      </aggregate>
      <flow>
        <min_rate>
          <lock>true</lock>
          <cap>0</cap>
        </min_rate>
        <policy>
          <default>fair</default>
          <allowed>fair</allowed>
        </policy>
        <target_rate>
          <default>700000</default>
          <cap>3000000000</cap>
        </target_rate>
      </flow>
    </bandwidth>
  </out>
</transfer>

Add the <max_mem> section in <scalekv> (may be called <scaledb> in older versions):

<scalekv>
  <sstore>
    <type>redis</type>
    <host>state_store_host</host>
    <port>state_store_port</port>
  </sstore>
  <baseport>43001</baseport>
  <max_mem>20000000000</max_mem>
  <!-- scalekv_max_mem: in bytes -->
</scalekv>

Set the Asperatrapd memory via firstboot. This is recommended when using c3.8xlarge images, to prevent trapd from consuming more memory than required.

# sh -c "sed -i 's/#system.buffer-pool.memory=-1/system.buffer-pool.memory=30GB/' /opt/aspera/etc/trapd/trap.properties"