[Firebird-devel] Some severe perfomance issue with (at least) FB 1.5 RC6 SS

[Jens <wei...@gm...>]

Hello to all fb developers,

i have a strange and curios problem with the FB 1.5 RC6 SS Release that makes it impossible to use it.

Unfortunatly i currently have no time to go into further investigation, but i think i schould tell you, that there _may_ be a big problem in the engine. I'll try to explain the problem as exact as i can.

On a customers site we use firebird as the database for their b2b web-shop application. We use it since sometime in 2001 beginning with FB 1.0/IB 6.0.

Some weeks ago we have updated the web-server including the firebird server (FB 1.5 RC3 was the first 1.5 Release we use). It work's fine so far. We've updated to FB 1.5 because of our overall positive experiences in other projects. 

On the application site we have not changed anything of the code, so it is the same clientside codebase as for the earlier releases of firebird.

With the current RC6 SS the engine behaves very strange. In the first few hours after starting the engine, everything works as expected. But at some point (i was'nt able to figure out what causes the problem) the engine begins thrashing. It makes amazing I/O read and write accesses with no need (no changes in website hits/s, no other special issues on the server). And it brings the server to a load of typical minimum 20 up to 80!!! Not because of cpu usage (cpu is 80 % idle) but because of the immense disk access. The only thing to stop this behaviour is to stop the engine and start it again. Then it works fine for a given amount of time and again begins thrashing after some time.

Firstly we used the FB 1.5 RC6 SS on a recent SuSE 8.2 Linux box. But we also tested it on FB 1.5 RC6 SS on a windows 2003 Server, so we think it is not a OS-specific problem.

A small example of what we call "thrashing": The used database is of about 2 GB in size. But approx 60 % of the database consists of BLOB data, that is not accessed very often. My last attempt with the win2003 server starts yesterday at 22:00. As i looked in the taskmanager today (13:00) the fbserver Process has cosumend about 103 GB(!) Read I/O and 105 GB(!) write I/O and used up to 142 Threads. But no one was on the web site (350 Page hits in that time period). And it was a freshly restored database, so there should be no garbage in it.

As i mentioned above, i have no time to investigate this problem. The website has to be back online asap. But i'll try to use the FB 1.5 RC6 CS on the linux box today, hoping that will solve the problem. If it does, i'll let you know.

Besides, one thing that might help you: On the linux box i've started the server with "fbmgr -start -forever". At the time of thrashing i noticed three running fbguard processes. AFAIU there should only one fbguard process running, shouldn't it?

cu
Jens