Helen Borrie wrote:

>At 04:26 PM 10/09/2004 +0300, you wrote:
>
>
>
>
>>Helen Borrie wrote:
>>
>>
>>
>>>At 03:16 PM 10/09/2004 +0300, you wrote:
>>>
>>>
>>>
>>>
>>>>Hi, all
>>>>
>>>>I have a production system runs with firebird about 3 years. About week
>>>>ago the system's behaviour becamed very strange. Server almost stops
>>>>responding after N-th user have connected to the database. Suddenly
>>>>simple queries becames too heavy - execution time raises from
>>>>milliseconds to minutes. When it happens CPU load is extreme high on
>>>>first hardware and moderate on second (hardware configurations are
>>>>described below), system memory is not exhausted and HDD activity is low.
>>>>
>>>>I've tried to check the database, backup/restore, upgrade Firebird from
>>>>1.5.0 to 1.5.1, switch from super to classic and back, migration to
>>>>another hardware - nothing helps. A few days ago server works before
>>>>hang about 3-4 hours. Now it takes a few minutes.
>>>>
>>>>The system serves about 15-25 connections. The database size is about
>>>>350-400Mb, last year added about 100Mb.
>>>>
>>>>Hardware: AMD Athlon XP3200+, 1.5Gb RAM, DPT Ultra-80 SCSI RAID 0 (3 x
>>>>9.1Gb HDD) + one standalone SCSI HDD for system boot, gigabit ethernet
>>>>controller, OS - Red Hat Linux 9.0 . I've tried another computer: Dual
>>>>Xeon 2.8 GHz, 2Gb RAM, 18Gb U160 SCSI HDD (10000RPM), gigabit ethernet
>>>>controller, OS - Fedora Core 1.
>>>>
>>>>I suspect the following:
>>>>- the production database and security.fdb are descendants from beta
>>>>versions of Firebird 1.5 or even release 1.0 . But backup/restore of it
>>>>was not helped and ODS version is 10.1 now.
>>>>- My old, but simple UDFs stops working, but how they did work about 3
>>>>last years? And there are no server crashes. By the way it worked on the
>>>>same system at least from 1.5.0 release date (April or May)
>>>>- some fundamental OS limits have been reached.
>>>>
>>>>And now the strangest thing: I've just replaced the HDD in the dual Xeon
>>>>system and installed the Firebird under W2003 Server there (affinity
>>>>mask was set to 15 in firebird.conf). The system is working excellent
>>>>under load already 4 hours.
>>>>
>>>>Please help me to diagnose and solve the problem.
>>>>
>>>>
>>>>
>>>>
>>>Clearly you didn't have Red Hat 9.1 or Fedora Core 1 for the past three
>>>years!!
>>>
>>>
>>>
>>Sure, but at least from RH9 release date. I've migrated from RH 7.2.
>>Fedora 1 was installed yesterday on fresh hardware.
>>
>>
>>
>>> Best guess here is that NPTL threading is messing things up. See
>>>the Linux installation notes in the Firebird 1.5 release notes for
>>>instructions on disabling the NTPTL; or install the NPTL kit for Firebird.
>>>
>>>
>>>
>>>
>>And how about the same problem with Super 1.5.0 and Classic 1.5.1 ?
>>
>>
>
>Sorry, I don't understand precisely what you are asking. But Superserver
>spawns a thread for each user process and for other processes, such as
>garbage collection; whereas Classic creates a unique process for each
>connection...so if you run the non-NPTL SS on a Linux that uses NPTL, then
>the creation of user processes is going to fall over.
>
>I really don't know whether NPTL raises issues for Classic at all. But,
>given that there is no NPTL kit for Classic, one can suppose not...
>
>./heLen
>
>

It would be great if someone helps me to diagnose problem: why any of
Firebird 1.5.x Linux distrubutions suddenly fail here in multiuser
environment. As I know the problem is out of scope of NPTL because it
was beginning with v1.5.0. It was working well before September and as I
know nothing was changed during last week or two. I've just tested
newest Firebird releases and another OS/hardware after the problem
appeared first time.

Thank you for your concern.

Andrew