maria-discuss team mailing list archive

[walter harms <wharms@xxxxxx>]

Am 31.01.2016 09:09, schrieb Sergei Golubchik:
> Hi, Reindl!
> 
> On Jan 30, Reindl Harald wrote:
>>
>>
>> Am 30.01.2016 um 21:07 schrieb walter harms:
>>> Aktualy I do now some profiling now we want to see the differences
>>> when switching 31-1.  We used myISAM since the biggest problem is
>>> speed and immoDB showed to be crash sensitive.  We store long time
>>> series data so the system is writing data all the time.
>>
>> for "writing data all the time" MyISAM is for sure a completly wrong 
>> decision because the performance strength of MyISAM was always on 
>> most-read workloads
>>
>> MyISAM *always* does a *complete table lock* for writes and don't allow 
>> concurrent writes without locking - that don't scale when you write all 
>> day long and there are table locks all day long
> 
> MyISAM should perform very good if inserts are *append only* (no updates
> or deletes). In this case MyISAM will not use an exclusive table lock
> and concurrent reads will be allowed.
> 
> It is typical for some kind of logging - one threads inserts *all the
> time* other threads are reading the data concurrently.
> 
> In fact, this is one of the use cases MyISAM was written for.
> 


This is was we actualy do with, never delete, very few updates and read is
no performance problem.

"crash sensitive" translates into: We had 3-4 occassions where an full backup was
needed, and all had all to do with immoDB tables broken beyond repair.


re,
 wh