In article <B9A2DDA03044D311BBD40008C75D69680B3719E4@dewdfx24>, 
[EMAIL PROTECTED] says...
> Hi,
> since SAPDB does not use any feature of a filesystem for its volumes(alias 
>devspaces) 
> any kind of filesystem overhead including a buffer cache is just an overhead. SAPDB 
>opens 
> using O_SYNC, but O_DIRECT is not yet used. All Read/Write activity use Systempages 
> which are aligned to 4KByte. The LINUX
> raw devices which are preferred are those 'unbuffered' /dev/raw/rawX.
> CU
> jrg

just tried using /dev/mdx (Linux SW RAID device) using RAID5, seems to be 
working. What should I expect in the therms of performance, compared with 
cooked FS? Any optimisation/configuration tips for RAID- strip sizes and 
similar?

(Yes I know hardware RAID is supposed to be faster, but this is no longer true 
- in most cases. Try >1 CPU over 2MHz plus LVD 160 controller, if you have more 
then 4 disks, spread over 2 SCSI busses. Now measure, and compare with HW 
controllers. Just don't compare it with controllers costing more and with more 
RAM then the whole server box, please - "let's be reasonable" (TM & famous last 
words).)

Hmmmm, just tried taking one disk down - all fine, CPU at 34% when doing full 
table select joining on fields without index...

Reinserted disk, same select on top of disk rebuilding had CPU at 13%. Rebuild 
of 9GM partition completed in 22 minutes, with database constantly doing full 
table selects with hash joins...

Looks good!

-- 
Yours, Andrej Falout, http://www.falout.com/disclaimer.html
Visit the OpenSource alternative, Aubit 4gl: http://aubit4gl.sourceforge.net
PLEASE NOTE: All HTML email sent to me WILL BE DELETED AUTOMATICALLY WITHOUT 
READING.


_______________________________________________
sapdb.general mailing list
[EMAIL PROTECTED]
http://listserv.sap.com/mailman/listinfo/sapdb.general

Reply via email to