Manjula G Kutty <[EMAIL PROTECTED]> writes:

> Knut Anders Hatlen wrote:
>
>>Manjula G Kutty <[EMAIL PROTECTED]> writes:
>>
>>  
>>
>>>Hi,
>>>
>>>I would like to know an estimate of howmany concurrent users are
>>>supported by derby network server. I know its hardware dependent, but
>>>for instance assuming a Windows box w/ 2Gb memory, single CPU, single
>>>7200rpm IDE drive.. Each user will be doing insert, delete, update and
>>>select.
>>>    
>>>
>>
>>What do you mean by supported? The number of concurrent connections
>>you can have without running out of memory? The number of concurrent
>>connection you can have with acceptable performance, for some
>>definition of acceptable? Something else?
>>
>>  
>>
> Hi
>
> Thanks for taking time to look into my mail.
>
> I would like to get the number of concurrent connections you can have
> with acceptable performance.

It's not possible to give an exact answer, since it depends on what
you consider acceptable, the type of load, the size and layout of your
tables, etc. However, to give an indication, I ran a test with up to
1000 concurrent connections. The attached graph shows the throughput
and average response time for an increasing number of connections.

The test clients operated on a 10 MB table (100 000 rows, 100 bytes
each). Each client executed one single-record operation (accessed by
primary key) per trasaction, of which 10% were insert operations, 10%
delete, 50% select and 30% update operations. The network server was
configured with 64 MB page cache, log and data were stored on separate
disks. Hardware configuration: 2x2.4GHz CPU, 2 GB RAM, 2 SCSI disks
(7200 rpm), disk write cache disabled. OS/JVM: Solaris 10 x86/Sun JVM
1.5.0. Derby version used was 10.2 (trunk).

The average response time for this mix of transactions is half a
second at 600 concurrent connections, one second at 850 clients and
1.5 seconds at about 950 clients. Maximum throughput is achieved at
about 150 connections.

Other configurations/workloads might give completely different
results, so YMMV.

-- 
Knut Anders

Attachment: performance.png
Description: PNG image

Reply via email to