Re: Question Regarding Slab Imbalance

2020-11-12 Thread Tony Wu
Yeah, default is 1.25. We can certainly start with 1.25. Thanks again, Tony Wu > On Nov 12, 2020, at 4:09 PM, dormando wrote: > > what's the default? you probably don't need to change it from the default > at all, which should have a good spread. default is 1.25 I think. > > On Thu, 12 Nov

Re: Question Regarding Slab Imbalance

2020-11-12 Thread dormando
what's the default? you probably don't need to change it from the default at all, which should have a good spread. default is 1.25 I think. On Thu, 12 Nov 2020, Tony Wu wrote: > Hi, > > The engine version we have running right now is 1.5.16. I just saw an > announcement that 1.6.6 is available,

Re: Question Regarding Slab Imbalance

2020-11-12 Thread Tony Wu
Hi, The engine version we have running right now is 1.5.16. I just saw an announcement that 1.6.6 is available, so we’ll probably plan to migrate. Adjusting growth factor makes sense. I did a quick google and don’t see any definitive formula to calculate an appropriate number. I’ll probably

Re: Question Regarding Slab Imbalance

2020-11-12 Thread dormando
1.5.what? also yes, chunk_size_growth_factor. don't set it to 1.02. set it to something so your slab classes are more evenly distributed. the max is 63. On Thu, 12 Nov 2020, Tony Wu wrote: > Hi Dormando, > > Thanks for the reply, I believe the engine version is 1.5. By multiplier > I assumed

Re: Question Regarding Slab Imbalance

2020-11-12 Thread Tony Wu
Hi Dormando, Thanks for the reply, I believe the engine version is 1.5. By multiplier I assumed you meant “chunk_size_growth_factor”, which is currently set at 1.02. Best, Tony Wu > On Nov 12, 2020, at 2:26 PM, dormando wrote: > > What version are they running now? That stat output looks

Re: Question Regarding Slab Imbalance

2020-11-12 Thread dormando
What version are they running now? That stat output looks pretty sparse. Unfortunately when it comes to elasticache the answer is probably to just beg them to upgrade to a newer version. they tend to run really old and newer verisons do a lot better at balancing memory. This is also suspect: >

Question Regarding Slab Imbalance

2020-11-12 Thread Tony Wu
We are currently using memcached provided by AWS ElastiCache service in modern mode to store web session keys among other things. We are observing session keys being evicted before TTL even though the cluster seems to have ample free memory, which hints at slab imbalance. I've included a print