Ben wrote:
> I also done some test in EMC SAN.
> If you only get the about 70MB/s for the SAN using dd.  I guess your
> LV stripe size is only 4K.  During my test using some disk IO
> performance tool, using 64K block size, I get the about 362MB/s
> throughput for random write IO.
>
> If 362MB/s, can I say the SAN configuration is ok?
>   
Not really. While your stripe size can influence things like READ, 
READNEXT (because it will trigger track read-ahead assuming you have 
told the file system to do this), you need to remember that the jBASE 
block size is 4K, not 64K and that unless your jBASE test program is 
performing only sequential reads, you will never get close to the 
performance of dd. In theory though a larger stripe will help some but 
not all programs. If the IO is totally random it might even slow things 
down as you may cause a read of a lot more data than you need.

What happens to your tests if you use a local disk and not the SAN?
> But during the T24 COB testing, the throughput is not good as I get
> using disk IO test tool even if I have increated the stripe size to
> 64K.
>   
It almost certainly is not performing sequential IO, hence the closest 
you will get to the iostone tests is the performance indicators given by 
a large (not a few megs but a lot of gigs) random read/write mix. 
Something like 6:1 read:write.
> Jim, please throw some light to my issue.
>
> We are using Jbase5.0.15. as database, T24 R8.003, and AIX 5.3.
>   
You can try to improve the performance by:

1) Defragmenting the jBASE files by jrf'ing them, or if they are being 
resized too big (which seems to be the case lately), then writing a 
script to dd them to new copies then moving them back over the original, 
using a large block size with dd (to give a large chance that the 
filesystem will allocate contiguous disk).
2) Make sure the files are sized correctly of course;
3) Take in to account the fact that there is probably transaction 
journaling going on;
4) Ask for advice from TEMENOS and the SAN supplier on tuning the SAN 
for jBASE access patterns;


First though, make your benchmarks with a local array, to give you a 
base point for comparison. Then make one change at a time and re-run the 
benchmarks.

Jim

--~--~---------~--~----~------------~-------~--~----~
Please read the posting guidelines at: 
http://groups.google.com/group/jBASE/web/Posting%20Guidelines

IMPORTANT: Type T24: at the start of the subject line for questions specific to 
Globus/T24

To post, send email to [email protected]
To unsubscribe, send email to [email protected]
For more options, visit this group at http://groups.google.com/group/jBASE?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to