Subject | Re: Firebird 2.1.3 Classic on Win2009 |
---|---|
Author | s3057043 |
Post date | 2010-11-21T23:05:10Z |
> You have avg hash len = 10. Double hash slots and avg hash len will be shorter two times, i.e. you can expect 5. I'd increase hash slots by 5-6 times for the start.Hello Group,
>
> And no, you don't need to restart service - full disconnect is enough. First connection will re-create lock table using new parameters.
>
> Regards,
> Vlad
>
We have now had the server running during peak usage with the following settings
LockMemSize = 6239027
LockHashSlots = 6007
The result has been a lower mutex wait but performance is still very slow and I think this is still to high.
---
LOCK_HEADER BLOCK
Version: 16, Active owner: 0, Length: 11121459, Used: 11111452
Semmask: 0x0, Flags: 0x0001
Enqs: 627479106, Converts: 367707, Rejects: 316416, Blocks: 1261345
Deadlock scans: 0, Deadlocks: 0, Scan interval: 10
Acquires: 650328654, Acquire blocks: 172022995, Spin count: 0
Mutex wait: 26.5%
Hash slots: 6007, Hash lengths (min/avg/max): 0/ 3/ 11
Remove node: 0, Insert queue: 0, Insert prior: 0
Owners (404): forward: 95488, backward: 10505952
Free owners (39): forward: 5047112, backward: 6457096
Free locks (2512): forward: 1360084, backward: 2805912
Free requests (2115): forward: 8937652, backward: 963852
Lock Ordering: Enabled
---
So we have boosted the values to
LockMemSize = 10356785
LockHashSlots = 10007
---
LOCK_HEADER BLOCK
Version: 16, Active owner: 0, Length: 10356785, Used: 6184060
Semmask: 0x0, Flags: 0x0001
Enqs: 52647655, Converts: 39481, Rejects: 43139, Blocks: 47160
Deadlock scans: 0, Deadlocks: 0, Scan interval: 10
Acquires: 54358501, Acquire blocks: 15237068, Spin count: 0
Mutex wait: 28.0%
Hash slots: 10007, Hash lengths (min/avg/max): 0/ 1/ 10
Remove node: 0, Insert queue: 0, Insert prior: 0
Owners (252): forward: 125324, backward: 5106132
Free owners (3): forward: 6165236, backward: 4836604
Free locks (763): forward: 3588628, backward: 2199812
Free requests (944): forward: 5089180, backward: 4334492
Lock Ordering: Enabled
---
If anything, the mutex wait seems worse for fewer owners?
What do these figures tell us?
Thanks in advance
Adam