For the extstore, it doesn't matter whether your data set is sharded or not 
as the scope is a single Memcached process. Note that data is not durable, 
if the Memcached process crashes, data can not be recovered.

Yes, ext_item_age=0 will write data to the disk quickly, leaving only keys 
+ meta data in memory. It's using FS cache as it does 
read/readv/pread/preadv.

If the probability of reading an item is high after it's written, you can 
increase the ext_item_age to keep it in memory. A possible downside of this 
is if your write rate is high, memory is full and the disk writes are slow, 
it will evict items before getting a chance to flush to the disk.

Extstore will work with both write-once or with updates.

More details are here: 
https://github.com/memcached/memcached/wiki/Extstore#tuning-guide. The 
original blog is here: https://memcached.org/blog/extstore-cloud/

On Friday, 4 December 2020 at 08:45:36 UTC-8 Jacob Eisinger wrote:

> Oh, and, one more important aspect.  Our data is mostly write-once, 
> read-many times.  There are no updates today.  And, in the future, there 
> will be very few updates.
>
> On Friday, December 4, 2020 at 11:05:18 AM UTC-5 Jacob Eisinger wrote:
>
>>
>> Howdy,
>>
>> We are investigating using Memcached's extstore as a 
>> single/non-sharded/non-clustered application cache of data larger than 
>> memory.  From my reading, it sounds like this is not the typical use case.  
>> Due to that, I wanted to see if there were any unintended consequences of 
>> our configuration.
>>
>> Specifically, we are looking to configure extstore with ext_item_age=0.  
>> From our initial testing, it looks like this means that this effectively 
>> turns Memcached into a mostly disk-based cache of values.  In particular, 
>> Memcached will attempt to write out pages as soon as they fill to disk.  
>> Then, when accessing values, the disk cache -> disk will be used.  Is that 
>> correct?
>>
>> We noticed the extstore file grows and the memcache memory stays 
>> relatively low.  If we wanted to keep more of the recent values in memory, 
>> should we just increase the ext_item_age to a couple of hours?  Are there 
>> unintended consequences of this setup?  Any other configuration that we 
>> should be looking at?
>>
>> Cheers,
>> Jacob
>>
>

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"memcached" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/memcached/6389b256-1204-47bc-b52e-5df3d63ef0a3n%40googlegroups.com.

Reply via email to