Already have 64GB…. (a few orders of magnitude over “Should be enough for anybody”…)
have tuned UDP quite a bit (sysctl) net.core.rmem_default=8388608 net.core.rmem_max=8388608 net.core.wmem_max=1048576 net.core.wmem_default=524288 net.ipv4.udp_rmem_min = 16384 net.ipv4.udp_wmem_min = 16384 Never see any problems with UDP or TCP packets, only TIPC. From: Erik Hugne [mailto:erik.hu...@gmail.com] Sent: Wednesday, July 20, 2016 3:05 PM To: Rune Torgersen Cc: tipc-discussion@lists.sourceforge.net Subject: RE: [tipc-discussion] ENOMEM No the buffer is shared across all protocols. Maybe try tuning the net.ipv4.udp_mem pressure/max? There are some tipc sysctls for memory use aswell.. Or.. buy more ram? ;) //E On Jul 20, 2016 7:37 PM, "Rune Torgersen" <ru...@innovsys.com<mailto:ru...@innovsys.com>> wrote: Makes some sense. Thing is we’re doing a LOT of UDP traffic at the same time and not much TIPC. (about 500Mbit/sec in and 1.2GBitsec out sustained video (1316 byte packets). The TIPC packets are larger than one MTU, so it might have fragmentation problems finding free SKB’s I guess. Is there a TIPC specific buffer pool (like there is for UDP and TCP)? From: Erik Hugne [mailto:erik.hu...@gmail.com<mailto:erik.hu...@gmail.com>] Sent: Wednesday, July 20, 2016 11:43 AM To: Rune Torgersen Cc: tipc-discussion@lists.sourceforge.net<mailto:tipc-discussion@lists.sourceforge.net> Subject: Re: [tipc-discussion] ENOMEM Each send() call will cause a new skb to be allocated from a cache. the caches are of fixed size, and do not grow (afaik, but the allocator can pick from higher order pool). The skbs are reclaimed by the cache when the stack is done with them. If the cache is exhausted when we call alloc_skb, userspace will get ENOMEM. i think you can check the cache fill level in /proc/slabinfo. other useful commands are slabtop and slabinfo //E On Jul 20, 2016 5:14 PM, "Rune Torgersen" <ru...@innovsys.com<mailto:ru...@innovsys.com>> wrote: Hi, just a simple question. What would make sendto() to a tipc RDM socket (destination being a local node address) (packet size 18000 to 40000 bytes) return a ENOMEM? ------------------------------------------------------------------------------ What NetFlow Analyzer can do for you? Monitors network bandwidth and traffic patterns at an interface-level. Reveals which users, apps, and protocols are consuming the most bandwidth. Provides multi-vendor support for NetFlow, J-Flow, sFlow and other flows. Make informed decisions using capacity planning reports.http://sdm.link/zohodev2dev _______________________________________________ tipc-discussion mailing list tipc-discussion@lists.sourceforge.net<mailto:tipc-discussion@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/tipc-discussion ------------------------------------------------------------------------------ What NetFlow Analyzer can do for you? Monitors network bandwidth and traffic patterns at an interface-level. Reveals which users, apps, and protocols are consuming the most bandwidth. Provides multi-vendor support for NetFlow, J-Flow, sFlow and other flows. Make informed decisions using capacity planning reports.http://sdm.link/zohodev2dev _______________________________________________ tipc-discussion mailing list tipc-discussion@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/tipc-discussion
