Differences

This shows you the differences between two versions of the page.

--- netflow:junipermx [2023/05/31 17:13] – veera
+++ netflow:junipermx [2023/06/21 16:50] (current) – [Default Flow table size] veera
@@ Line 65: / Line 65: @@
+===== Default Flow table size =====
+By default MX has a very small flow table size of 1K, this can be observed by Trisul as a very slow netflow records / second rate. Go to "Netflow > Flow Sources"  then see the Flow records /sec.  This happens usually when you are configuring JFlow on a Juniper MX for the first time and are using the default 1K flow table size. Juniper recommends configuring the ipv4-flow-table-size to 4 that means 1M for most cases ((https://www.juniper.net/documentation/us/en/software/junos/flow-monitoring/topics/ref/statement/ipv4-flow-table-size.html))
+Use the following command
+<code bash>
+set chassis fpc 0 inline-services flow-table-size ipv4-flow-table-size 4
+set chassis fpc 0 inline-services flow-table-size ipv6-flow-table-size 1
+</code>
+The flow table is used by the FPC to do JFlow (Netflow), due to the low default limit of 1K, the table fills up quickly and most of the new flows are not able to be counted. This results in very low JFlow/Netflow rate because in large networks long lived flows never leave the table and are only subject to active flow timeout.
+You can also try to use flex-flow-sizing , this is supposed to automatically scale the flow table. But there are some reports online it does not work as expected on MX204 ((https://phabricator.wikimedia.org/T248394))
+Perhaps at some point Juniper MX will release with sensible default for  ''ipv4-flow-table-size'' parameter
+====== Useful debugging ======
+Use the flow errors status commands
+<code bash>
+run show services accounting flow inline-jflow fpc-slot 0
+run show services accounting errors inline-jflow fpc-slot 0
+run show services accounting status inline-jflow fpc-slot 0
+</code>