On Apr 19, 2011, at 10:52 AM, Andrew Hutchings wrote:

> On Tue, 2011-04-19 at 09:06 -0500, Tim Soderstrom wrote:
>> On Apr 19, 2011, at 8:55 AM, Olaf van der Spek wrote:
>> 
>>> On Tue, Apr 19, 2011 at 3:41 PM, Tim Soderstrom
>>> <[email protected]> wrote:
>>>> Perhaps I am misunderstanding you, but would a transaction not be what 
>>>> you're asking for?
>>>> 
>>>> BEGIN;
>>>> UPDATE ... ;
>>>> UPDATE ... ;
>>>> ...
>>>> COMMIT;
>>> 
>>> Eh, no. That'd still be 10000 update statements.
>> 
>> Yes, but the writes will occur in a single transaction when you COMMIT so it 
>> should be faster than doing isolated UPDATEs. Use a prepared statement if 
>> you want to get even more speed out of it (though I suspect any speed 
>> improvements may be sort of marginal, and that's assumption not bounded by 
>> any particular benchmarks I have done that show differences either way).
> 
> Agreed, the parse/optimize time for 10000 key/value statements in a
> transaction is almost unmeasurable in Drizzle.  When you get into the
> millions I think there is a small measurable overhead.
> 
>>>> That is, if you wanted to have all those statements go through at the same 
>>>> time. I think if you wanted to improve speeds of updates, you may want to 
>>>> look at PBMS, the BLOB streaming stuff, which functions similar to 
>>>> HandlerSocket. I'm not sure if HandlerSocket can be built into Drizzle 
>>>> but, last I remember, PBMS is already there ready to go as a plugin.
>>> 
>>> This isn't about blobs. How would PBMS help?
>> 
>> It's a misnomer that PBMS handles just BLOBs. Like HandlerSocket, it can 
>> deal with pretty much any sort of data as far as I'm aware. Paul wrote a 
>> post about this:
>> 
>> http://pbxt.blogspot.com/2010/12/handlersocket-why-did-out-version-did.html
>> 
>> It's faster because you don't use the SQL optimizer, though I don't think 
>> you can do transactional updates with it. Still, if it's anything close to 
>> the speed of HandlerSocket, I would take a look at this since you're doing a 
>> bunch of key-value updates anyway (hence you don't need a SQL layer for 
>> those really).
> 
> I was actually discussing NoSQL access with Brian at the Drizzle
> Developer Day.  I like his idea of using a memcached style access.  The
> apps I have written for companies use memcached anyway, so it would be
> pretty trivial to convert these.

Oooh that would be actually very very cool!
_______________________________________________
Mailing list: https://launchpad.net/~drizzle-discuss
Post to     : [email protected]
Unsubscribe : https://launchpad.net/~drizzle-discuss
More help   : https://help.launchpad.net/ListHelp

Reply via email to