Cache activity vs Page faults on Wired tiger

1,046 views

Skip to first unread message

Joe Waller

unread,

May 4, 2016, 11:39:05 PM5/4/16

to mongod...@googlegroups.com

Hi all,

I am seeing quite a bit of "Read into" activity on the cache but my page faults are 0. Shouldn't the "Read into" activity of the cache be captured as page faults?

thanks

Joe

Kevin Adistambha

unread,

May 11, 2016, 12:48:16 AM5/11/16

to mongodb-user

Hi Joe,

I am seeing quite a bit of “Read into” activity on the cache but my page faults are 0. Shouldn’t the “Read into” activity of the cache be captured as page faults?

Where are you seeing these statistics? Did you gather them from db.serverStatus() command?

Page fault is MMAPv1 specific statistics that tracks how often MongoDB must access the disk when the associated data was not loaded into memory yet. Historically, this metric was used to check whether there is physical memory pressure in the system. This statistics is less relevant for the WiredTiger storage engine, since WiredTiger tracks its statistics differently.

In WiredTiger, there is no single “page fault”-like metric that represents memory usage. The relevant statistics in WiredTiger are the wiredTiger.cache document from the output of db.serverStatus(). In particular, wiredTiger.cache.unmodified pages evicted shows the number of pages evicted from the cache. If it corresponds to an increase in wiredTiger.cache.pages read into cache, it may indicate that pages are evicted from the cache to be replaced by another page read from disk. Possible reason include large collection scans.

Please see the wiredTiger.cache page for more information regarding this topic.