[rbldnsd] Possible Small Bug

[Michael Tokarev]

Jeff Chan wrote:
> On Friday, January 28, 2005, 9:27:51 AM, Jon Lewis wrote:
[]
>>But...I'm seeing PII/PIII Celeron 400-500mhz systems handling >1300qps and
>>rbldnsd keeps on trucking.
> 
> Thanks for the benchmark; most of our own servers are a bit
> faster, and we're doing about 50 qps, so it's good to know we
> have some headroom. 

Speaking of "benchmarks" -- this two graphs:
  http://www.spambusters.org.ar/nsg/w_q.gif
  http://www.spambusters.org.ar/nsg/sbl/w_q.gif
shows weekly stats for two instances of rbldnsd running on single
PIII-600 machine.  This is j.ns.dsbl.org aka ns6-cbl.abuseat.org
aka t.ns.spamhaus.org.  2.5 + 1.9 = 4.4 qps average load.  Top:

top - 23:58:26 up 85 days,  3:47,  1 user,  load average: 0.69, 0.55, 0.49
Tasks:  48 total,   4 running,  44 sleeping,   0 stopped,   0 zombie
Cpu(s):   6.0% user,  18.2% system,   0.0% nice,  75.8% idle
Mem:    255888k total,   224540k used,    31348k free,    20532k buffers
Swap:   499960k total,    65436k used,   434524k free,   120732k cached

   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
28660 rbldns    19   0 50924  49m  300 R  9.0 19.8   5654:27 rbldnsd
28663 rbldns    14   0 15056  14m  280 R  7.3  5.8   4290:03 rbldnsd

Note the system vs user time: most time is spent in system code, I guess
it's the network packets and NIC interrupts handling...  Typical vmstat
output:

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
  r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
  2  0  66380  82504   7040  83660    0    0     0    26 11143 16906  5 25 70  0
  2  0  66380  82504   7040  83660    0    0     0     0 11182 16770  3 28 69  0
  3  1  66380  82504   7040  83660    0    0     0    23 11194 17031  6 27 67  0
  2  0  66380  82500   7044  83660    0    0     0     1 11133 16989  4 28 67  0
  2  0  66380  82500   7044  83660    0    0     0     0 10968 16724  4 26 69  0
  2  0  66380  82496   7048  83660    0    0     0     2 11264 17253  4 29 67  0
  2  0  66380  82496   7048  83660    0    0     0     0 11137 16845  4 28 67  0

11K interrupts/sec and 16K context switches/sec is quite a few for 600MHz system.. ;)
(The main interrupt source is the nic - tulip driver and, according to lspci,
00:0a.0 Ethernet controller: Macronix, Inc. [MXIC] MX987x5 (rev 25) nic).

/mjt