Page 2 of 2
Posted: Mon Oct 22, 2012 6:28 am
by chulett
Well... allegedly it runs fine on one server and not the other. I've seen that when the 'works fine' one is 64bit and the 'not so fine' one is 32bit. Any chance of that?
Posted: Tue Oct 23, 2012 2:45 am
by tehavele
Yes, u are right. But my greater concern is that how can the same 7 Gb of data gets referenced in one server in same job and fails in another !!!
Strange.
Posted: Tue Oct 23, 2012 6:38 am
by chulett
I know I'm right in a general sense, can you confirm or deny if that applies to your specific case? I was asking if your servers differ in their "bitness"... do they?
Posted: Sun Oct 28, 2012 8:37 pm
by tehavele
Oh sorry chullet, I referred to the ArndW comments earlier.
Well chullet, in 'bitness' the servers are similar 64Bit. Infact in the same server I am now able to run this job completely with a EV "APT_LUTCREATE_MMAP = 1". This changes the way job uses the memory [Work arround]. Thus I suspect something wrong with the basic memory Utilization.
I will post the details of both the server again for your analysis. We need to find whats wrong with this(Slow) server:
Fast server CPU info:
-----------------------------------
1) 2 Intel(R) Itanium 2 9100 series processors (1.59 GHz, 12 MB)
266 MHz bus, CPU version A1
2) Memory: 16378 MB (15.99 GB)
3) OS info:
Nodename: cdhiiad1
Release: HP-UX B.11.31
Version: U (unlimited-user license)
Machine: ia64
ID Number: 1789613992
Slow server CPU info:
-----------------------------------
1) 4 Intel(R) Itanium(R) Processor 9350s (1.73 GHz, 6 MB)
2.39 GT/s QPI, CPU version E0
2) Memory: 24570 MB (23.99 GB)
3) OS info:
Nodename: cqhiiad1
Release: HP-UX B.11.31
Version: U (unlimited-user license)
Machine: ia64
ID Number: 0076437800
Slower server has better resources than the fast server[Except memory marked in red] and very same OS.
Posted: Mon Oct 29, 2012 6:44 am
by ArndW
Have you actually checked how much data is being used for the lookup reference set on the machine that is working?
Posted: Wed Oct 31, 2012 2:15 am
by tehavele
Hi ArndW
Considering the fact that we can only compare two servers' performances with identical data only, I have used the same Input and same Reference data for both servers.
Clearly, servers fair differently on the same load.
Posted: Wed Oct 31, 2012 4:01 am
by ArndW
Very strange, I cannot see how 7Gb can be used in a lookup stage. I wonder if the 7Gb might be the full table size but that the lookup only uses keys and a select subset of the data columns? Also, the "APT_LUTCREATE_MMAP = 1" setting applies to using lookup filesets but you stated that the lookup reference data is coming from a table.