Re: [Fwd: Re: Lots of Interrupts when accessing a NFS share]

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

(I don't subscribe this list so please Cc: me)

Well, I'm not sure if it's a problem on the driver or chip,
but as you can see in a comment in src/sys/dev/ic/rtl81x9.c
http://cvsweb.netbsd.org/bsdweb.cgi/~checkout~/src/sys/dev/ic/rtl81x9.c?rev=1.81
the Realtek 8139 is not so smart.
On the other hand, modern CPU is so overmuscled
to handle interrupt storm from such dumb device.

Anyway, ~10000 intrs/sec is not surpising for me
because find(1) may send/receive a bunch of small packets.

My Duron 1300MHz PC also shows the similar output on find(1)
over NFS via rtk(4):
---
    2 users    Load  0.56  0.21  0.08                  Fri Mar 14 01:04:03

Proc:r  d  s  w     Csw    Trp    Sys   Int   Sof    Flt      PAGING   SWAPPING
          13      11559     18  36291 11649  5919      5      in  out   in  out
                                                        ops
  19.4% Sy   3.4% Us   0.0% Ni  15.6% In  61.6% Id    pages
|    |    |    |    |    |    |    |    |    |    |
==========>%%%%%%%%                                                       forks
                                                                          fkppw
           memory totals (in kB)           17555 Interrupts               fksvm
          real  virtual     free            5804 softint net/0            pwait
Active   36524    36524   558764               2 softint bio/0            relck
All     203320   203320  1566760                 softi bio block          rlkok
                                             100 softint clk/0            noram
Namei         Sys-cache     Proc-cache         2 pic0 pin 14              ndcpy
    Calls     hits    %     hits     %           pic0 pin 15              fltcp
    10308       22    0                    11552 pic0 pin 10            5 zfod
                                                 pic0 pin 6               cow
Disks:   fd0   md0   wd0   cd0  nfs0          95 pic0 pin 0            64 fmin
 seeks                                                                 85 ftarg
 xfers                 2                                                  itarg
 bytes              4710                                             1885 wired
 %busy               0.2                                                  pdfre
                                                                          pdscn
---
but re(4) (RTL8169) and ex(4) (3C905B) on the same machine
also show the similar numbers.

I'd suggest to use proper benchmark programs (like ttcp or netperf)
to measure performance. A number of interrupts is not so important
if (overmuscled) CPU is still idle, but bandwidth on the bus-master
device may limit network throughput.
---
Izumi Tsutsui