Subject | Re: [Firebird-general] Data files on raw devices |
---|---|
Author | Dmitry Kuzmenko |
Post date | 2004-10-11T06:35:25Z |
Hello, manoel_sa!
Thursday, September 23, 2004, 6:07:02 PM, you wrote:
m> There is a compile option in Firebird 1.5 (--enable-raw-devices) for
m> raw devices. Does this mean we may have data on raw devices (on linux)
m> ? Are there any examples. I haven't seen any reference to this in the
m> docs.
In 1999 I was trained administering Informix with SAP/R3.
I've asked trainer - if there any performance gain with
databases on raw devices. She said, that even she knows
some cases when it was used, never heard of any performance
increase (Oracle, Informix).
Next years I've heard other times that databases on raw
devices has no advantage in performance over filesystems.
As I understand correctly, raw device will differ from fs
only by not using operating system cache. This can be now
done in Yaffil (win32_disable_file_cache), but it produces
strange results, mostly depending on fs cluster and page size
correlation.
--
Dmitri Kouzmenko
Thursday, September 23, 2004, 6:07:02 PM, you wrote:
m> There is a compile option in Firebird 1.5 (--enable-raw-devices) for
m> raw devices. Does this mean we may have data on raw devices (on linux)
m> ? Are there any examples. I haven't seen any reference to this in the
m> docs.
In 1999 I was trained administering Informix with SAP/R3.
I've asked trainer - if there any performance gain with
databases on raw devices. She said, that even she knows
some cases when it was used, never heard of any performance
increase (Oracle, Informix).
Next years I've heard other times that databases on raw
devices has no advantage in performance over filesystems.
As I understand correctly, raw device will differ from fs
only by not using operating system cache. This can be now
done in Yaffil (win32_disable_file_cache), but it produces
strange results, mostly depending on fs cluster and page size
correlation.
--
Dmitri Kouzmenko