1

Topic: hugepages that happens at their exhaustion

There is a copy of Oracle 11.2, works under RHEL. It is configured for operation with hugepages. All works, but here there is one parameter in system which guards - hugepages_free. In the course of operation it gradually "thaws". During working off of heavy requests in a DBMS it "thaws" faster.
What happens, if hugepages_free will be settled, becomes equal 0? the Positive dynamics on a reuse of already selected storage it is not visible. And here constant reduction of this parameter says that at some instant possibility receive the next page.

2

Re: hugepages that happens at their exhaustion

helgisbox;
huge pages it is reserved the fixed volume at copy start.

3

Re: hugepages that happens at their exhaustion

helgisbox wrote:

In the course of operation it gradually "thaws"

What exactly "thaws"? What speaks "HugePages_Rsvd"?

[oracle@srv1 ~] $ grep Huge/proc/meminfo
AnonHugePages: 5044224 kB
HugePages_Total: 209715
HugePages_Free: 23895
HugePages_Rsvd: 17641
HugePages_Surp: 0
Hugepagesize: 2048 kB

4

Re: hugepages that happens at their exhaustion

Takurava,

HugePages_Total: 8925
HugePages_Free: 3105
HugePages_Rsvd: 3101

5

Re: hugepages that happens at their exhaustion

helgisbox, Reserved together with Free decreases. The difference is in between equal 4, a constant.

6

Re: hugepages that happens at their exhaustion

helgisbox wrote:

helgisbox, Reserved together with Free decreases. The difference is in between equal 4, a constant.

there were no thoughts?
The basis reserved under itself N pages - while they are considered free, but as required the basis to itself takes away them.
If free hardly it is more than reserdev all apprx. If one in one - it is necessary to look in alert.log that there:
11 - Large Pages configured> Large Pages used (Total Shared Global Region in Large Pages = 100 %)

************************ Large Pages Information *******************
Per process system memlock (soft) limit = 238 GB
Total Shared Global Region in Large Pages = 10 GB (100 %)
Large Pages used by this instance: 5233 (10 GB)
Large Pages unused system wide = 0 (0 KB)
Large Pages configured system wide = 104000 (203 GB)
Large Page size = 2048 KB

12 - AVAILABLE_PAGES> EXPECTED_PAGES

Wed Dec 16 2:44:58 PM 2015
PAGESIZE AVAILABLE_PAGES EXPECTED_PAGES ALLOCATED_PAGES ERROR (s)
Wed Dec 16 2:44:58 PM 2015
4K Configured 7 7 NONE
Wed Dec 16 2:44:58 PM 2015
2048K 51542 5217 5217 NONE

7

Re: hugepages that happens at their exhaustion

Frightens a monotonicity . On idea should be what  "reusage". That is occupied, for example 5000  from 8100 reserved, 3000, 7900. And here like as it works gradually reaching a limit. What will be, if approaches closely to number reserved, there will be no error and the copy when the next attempt to receive a portion of storage will not be completed by success does not stop?
There are any representations internal (can in x $ tables) where it is in more details painted and it will be clear, what it not a monotonicity in growth hugepages?
If there is a link to manuals, ask to throw in me.

8

Re: hugepages that happens at their exhaustion

helgisbox wrote:

when the next attempt to receive a portion of storage will not be completed by success?

Attempt to receive storage crowned success at copy start, it is more than attempts will not receive storage, only attempts will use it . And . But not to release.

9

Re: hugepages that happens at their exhaustion

Passed by...;
In the copy as it is possible to see how many pages already upon now it is used, and how many it is free? From the point of view of OS they are reserved and will continue to be eaten off as required. From the point of view of OS the reserve gradually understands - here there is no crime. And here that the copy with it storage does,  to learn? The monotonous growth is a result of a fragmentation?

10

Re: hugepages that happens at their exhaustion

helgisbox;
Memory Architecture