[Rpm-maint] Scalability of RPM

[Christoph Wissing]

Dear RPM experts,

I have a potential problem with the scalability of RPM. Let me briefly describe the use case. RPM is used to manage the software of the CMS-Experiment (http://cmsinfo.cern.ch/outreach/). Some might have read about the creation of black whole in context of these experiments (Rubbish!). But what's the problem with RPM? At some point, when quite some software is installed there are problems with the RPM database and all rpm commands that want to install new software stop with this error:

memory alloc (4 bytes) returned NULL. 

The number can be different, but 4 is rather pupular. Often it helps to remove some software, but some even the erase command fails with the same error.

The problem only appears on 32-bit (Linux) plattforms, the most used distibution is Scientific Linux 4, a recompiled RedHat Enterprise like CentOS. Some investigations have shown that the rpm command allocates more and more memory and when it reaches about 2GB it crashes. Running on 64-bit shows actually the same (very high) memory usage, but the task seems not to crash, most like because the system can just allocate more memory for a task.

The question is if there can be done something about it and if there some limitations known. The number of files that are actually stored are quite big, some millions easyly. But on the other hand installing a complete Linux distribution with several kernels and headers show end up with a number of files in the same order of magnitude.

Can someone from you help? What additional information do you need? The databases are a bit large (200MB) to attach them to this mail.

Any help is very welcome!

Thanks a lot in advance and best wishes,

Christoph

-- 
+-----------------------------------+
| Christoph Wissing       DESY - IT |
| E-Mail: christoph.wissing at desy.de |
| Phone:  +49(0)40/8998-4122        |
+-----------------------------------+