Hi,
I have NSD installations on a number of systems with different hardware
architectures. On some, like OS X on PPC, the performance of zonec is
excellent. But, on my UltraSparc II production systems zonec is really
slow.
Machine PBG4 12" Netra T1 DC200
CPU PPC 1.25Ghz US II 500MHz
OS X 10.3.9 Sol 8 64-bit
Time 1min 57sec 23 min.
At no time did the machines swap and they were fairly unloaded, the mac
less so if anything. There is one thing that could differ additionally; the
Suns run Solaris in 64-bit mode and the NSD installation is a 64-bit one;
due to OpenSSL being 64bit.
I think there is something strange going on here; we should not see a
12-fold loss in performance from going to the Sun. Any thoughts?
[On 27 Aug, @18:45, M??ns Nilsson wrote in "Performance of zonec with dnss ..."]
At no time did the machines swap and they were fairly unloaded, the mac
less so if anything. There is one thing that could differ additionally; the
Suns run Solaris in 64-bit mode and the NSD installation is a 64-bit one;
due to OpenSSL being 64bit.
I think there is something strange going on here; we should not see a
12-fold loss in performance from going to the Sun. Any thoughts?
Nope, very strange indeed.
Would it be possible to get some hard data? Like a gprof profile or
something simular?
Yes, but I might need some concrete instructions on procedure to get the
data.
Måns Nilsson wrote:
Yes, but I might need some concrete instructions on procedure to get the
data.
Assuming you're using GCC and have gprof available, configure with CFLAGS='-O2 -g -pg', make clean and recompile.
Then run zonec. A gmon.out or zonec.gmon file should be generated. Use 'gprof zonec gmon.out' or 'gprof zonec zonec.gmon'.
It would be useful if you could do this both on PPC and SPARC, that way we may be able to see if any part of the coding is performing much worse on SPARC than on PPC.
Also, what is the performance difference without DNSSEC? A factor of ~11.5 as well?
Erik