libre #smartos

18 Mar 2026

SmartOS -- A Modern, Open Source, Datacenter Operating System | smartos.org | github.com/TritonDataCenter/smartos-live | github.com/TritonDataCenter/triton | Security -- TritonDataCenter.com/security-issues | Channel logs - log.omnios.org/smartos

• 13:29
  huy
  Hello, I treated my microserver gen8 with a x540-t2 card but I can't push more than 670Mbps, so it's no better with ixgbe than with bge. Could the bottleneck be the cpu? It's a E3-1220L V2.
• 13:34
  jbk
  could be a number of things.. probably the first question is how are you measuring that?
• 13:35
  jbk
  generally you should be able to get at least a gigabit on a single connection though..
• 13:37
  jbk
  once you get past that, it often requires multiple connections (i.e. parallelism) to get the higher speeds
• 14:21
  huy
  jbk : iperf3 -c ping.online.net
• 15:59
  sommerfeld
  huy: what's your round trip time (as reported by ping -s) to that host?
• 15:59
  sommerfeld
  and have you run iperf3 over LAN to a nearby system?
• 16:02
  sommerfeld
  you may be limited by TCP send window sizing if it's a WAN speed test; you can adjust those values upward with ipadm but overdoing it can hurt your LAN performance (this is a bug, and I'm working on it..)
• 16:04
  sommerfeld
  (rule of thumb: the buffer size must exceed the bandwidth-delay product of the path. delay= round trip time; bandwidth: speed of your share of the slowest link (in bytes per second).
• 16:05
  huy
  PING ping.online.net: 56 data bytes
• 16:05
  huy
  64 bytes from 51.158.1.21: icmp_seq=0. time=14.372 ms
• 16:06
  sommerfeld
  destination 15ms away, 1Gbps link -> BDP is 0.015 seconds * 1Gbps = 15000000bits = 1875000 bytes
• 16:06
  sommerfeld
  what does "ipadm show-prop tcp" show?
• 16:07
  sommerfeld
  specifically the max_buf and send_buf properties?
• 16:07
  huy
  tcp max_buf rw 1048576 -- 1048576 8192-1073741824
• 16:08
  huy
  tcp send_buf rw 128000 -- 128000 4096-1048576
• 16:08
  sommerfeld
  bump up send_buf a bit and see if it gets better.
• 16:08
  sommerfeld
  and maybe max_buf as well.
• 16:14
  huy
  Hm it does not change much, it's actually a bit worse, I changed send_buf to 164000 and max_buf to 2097152
• 16:15
  sommerfeld
  you may be at the limits of that path.
• 16:20
  sommerfeld
  do you have a local (wired ethernet LAN) host to test against?
• 16:21
  huy
  I'm trying that now
• 16:28
  huy
  927 Mbits/sec sender, 925 Mbits/sec receiver
• 16:29
  huy
  The other end only as 1Gbps interface
• 16:33
  sommerfeld
  after some tuning I see up to 9.39 Gbits/sec between two systems with X710 (i40e) connected to the same switch.
• 16:34
  danmcd
  A couple of things to remember:
• 16:34
  danmcd
  - those ipadm properties are *per zone* in SmartOS Native-or-LX zones. Fixing it in the GZ doesn't affect zones at all.
• 16:35
  danmcd
  - Actually, the other thing was already covered (what's the path between you and test-peer).
• 16:36
  huy
  It's a laptop running debian. The thing is that I get the same numbers to ping.online.net when I connect it to the same uplink, which is a ffth box.
• 16:36
  huy
  And I was testing in the GZ