+ Reply to Thread
Results 1 to 2 of 2
Thread: Memcache hit rate & suggestion
03-17-2012 12:22 PM #1
- Join Date
- Apr 2011
Memcache hit rate & suggestion
first with the suggestion: could we have a "server" section where we can ask specific setup questions and maybe have posts there where Brian brags about the server setup Prosperent has and what they're doing/how they doing it?
Next is Memcache. I have 2 sites running on different cloud servers, and figured I could speed things up (which, by the looks of it does) adding a 3rd Memcache only server with 1 gb ram as a start (so that I don't have to upgrade both instances for more ram)
Because I'm on the same cloud network, the response time is less than 1ms which is nice, my cache is sitting at 90% full at the moment (768mb dedicated to memcache) but the hit ratio is only 11%
This is making me wonder if investing in memcache is going to be financially viable (obviously the more sites I add on the more use I get out of it), but it also makes me wonder if there's any specific tips you guys might have to increase the hit ratio to about 60-70%?
03-17-2012 02:21 PM #2
I definitely think we need a server forum and will get one setup.
On the cache front, it's always difficult to predict. There are so many query variations that having a high hit rate can take quite a bit of memory. I have tested as high as 96gb to get us over 60 percent, but noticed that as low as 16gb we are in the 50 percent range. I think the new description optional queries will go a long way toward helping. An API result without descriptions is literally 1/10the the size of a standard query, and with good site design can be used on 90+ percent of your pages. On thenetpool there are only two pages that use descriptions. Our list view search results ( which we will modify to not use descriptions), and our product detail pages. These pages only request details for a single product though so the response is still tiny. At the end of the day it is going to come down to smart site designs that only rely on necessary data and cache smaller repossess more efficiently.