| Subject | Re: [firebird-support] Write into a .fdb file without the overhead of API's |
|---|---|
| Author | David Johnson |
| Post date | 2005-07-02T02:51:59Z |
On Fri, 2005-07-01 at 11:39 +0530, Revathy Mahalingam wrote:
cache RAM is on your 10Gbit fiber channel DASD?
On an 800 MHz Intel desktop with standard IDE disk drive, I achieved a
steady 12,000,000 rows per hour (3,330 rows per second more or less) for
insertions using the classic model server. On real server hardware,
this leaves you with some room for other work to happen.
Make sure you attack the right bottleneck before tackling the API.
While there are inefficiencies in the wire protocol and passing buffers
between layers, those are on the order of nanoseconds, whereas the IO
limits of the hard drive system are measured in 10's of milliseconds.
If you save 1% of your time in skipping the API (at the expense of a
couple of man-years of work, plus the support that goes with it), and
never address the 99% of the time that the disk system is holding up the
works, you have only wasted your time and your customers' money.
> Hi,How many thousand rows per second performance do you need, and how much
>
> One of our clients has some existing products, which generate real-time
> data. Now they intend to integrate a backend for these products.
>
> Can anyone tell me what is the raw format of the Firebird database files, so
> that I can directly write into a file without the overhead of the API's
>
> Thanks n regards
> Revathy
>
>
cache RAM is on your 10Gbit fiber channel DASD?
On an 800 MHz Intel desktop with standard IDE disk drive, I achieved a
steady 12,000,000 rows per hour (3,330 rows per second more or less) for
insertions using the classic model server. On real server hardware,
this leaves you with some room for other work to happen.
Make sure you attack the right bottleneck before tackling the API.
While there are inefficiencies in the wire protocol and passing buffers
between layers, those are on the order of nanoseconds, whereas the IO
limits of the hard drive system are measured in 10's of milliseconds.
If you save 1% of your time in skipping the API (at the expense of a
couple of man-years of work, plus the support that goes with it), and
never address the 99% of the time that the disk system is holding up the
works, you have only wasted your time and your customers' money.