Actually, those IPC calls are a problem in general on the INTEL platform. Something like 200 instructions cycles are
requiredfor each one.
On a sun SPARC, with it's windowed registers, the average comes down to something like 10 cycles.
Hopefully, on this new 64 bit archtecture that intel is introducing, the one the emulates the old PC architecture
insteadof implementing it again, they have absorbed all the last 15 years of advances in Server Processors and
COMPLETELYthrow away all the old INTEL architecture baggage.
Dennis Gearon wrote:
> IPC, 'interrupt procedure call'?
>
> scott.marlowe wrote:
>
>> On Thu, 17 Jul 2003, Markus Heinz wrote:
>>
>>
>>> Hello,
>>>
>>> i'm evaluating Postgresql on win32 (winnt, win2k). I'm using a small
>>> java benchmark from FirstSQL (see attachments).
>>> I've ran the benchmark on three pc's
>>> 1) Pentium III 750 MHz Winnt Sp6a, 372 MB PC100 RAM, IBM 26GB HD
>>> 7200RPM
>>> 2) Pentium 4 1.4GHz Win2k SP4, 256MB RRAM, WD 20GB HD 7200RPM
>>> 3) Athlon XP 2400+ Win2k SP4, 512 DDR 266, Seagate 120GB HD 7200RPM
>>>
>>> using identical cygwin and PeerDirect configs on all PCs.
>>> To my surprise the Pentium III on Winnt was three times faster than
>>> the Pentium IV 1.4 GHz
>>> and two times faster than the Athlon XP 2400+.
>>> Are there known problems with postgressql and Win2k SP4 ?
>>
>>
>>
>> No, Microsoft decided to basically change the preferred method for
>> applications to talk to each other with the release of win2k. while
>> Winnt had fast IPC, (the IPC that cygwin uses relies on it) 2k had
>> much slower stock IPC. There are a few articles floating around the
>> net about it. It's one of the reasons many older NT servers are still
>> running, because the software running on them will be slower on
>> machines running 2k and above.
>>
>>
>> ---------------------------(end of broadcast)---------------------------
>> TIP 3: if posting/reading through Usenet, please send an appropriate
>> subscribe-nomail command to majordomo@postgresql.org so that your
>> message can get through to the mailing list cleanly
>>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 3: if posting/reading through Usenet, please send an appropriate
> subscribe-nomail command to majordomo@postgresql.org so that your
> message can get through to the mailing list cleanly
>