On December 13, 2018 1:47:44 AM GMT+08:00, Quanah Gibson-Mount <quanah@sym= as=2Ecom> wrote:
Hi,
Thanks for the report! I have some questions:
a) Did you test with writemap enabled for mdb?
No, but I did test the unpatched slapd with writemap=2E it was not much fa= ster=2E I can test the patch with writemap to see what happens=2E
b) Did you test the effect on other operating systems? (I=2Ee=2E, Windows=
,=20
Solaris, FreeBSD, etc)?
No , I dont have any=2E The only thing I have is a x86-64 pc with linux=2E= I did try to avoid touching any os specific codes
c) Did you test multiple filesystems under Linux(I=2Ee=2E, ext2, ext3, ext4,=20 xfs, etc)?
no=2E I only tested with ext4=2E
d) What filesystem did you use when testing? Was journaling enabled or
disabled? What was the default flush behavior of the filesystem you used?
I used nobarrier, data=3Dwriteback mount options, which are faster than th= e default, but not significantly=2E I am using a very fast SSD though=2E
I=2Ee=2E, if you set dbnosync at the mdb level, the underlying filesystem flush=20 behavior will then be in effect (with ext4, for example, this defaults to 5=20 seconds)
something like that=2E it also depends on vm setting, like dirty ratio=2E
e) Did you test when the database is also writing to an underlying=20 accesslog database that is also backed by mdb?
no, I dont konw how=2E
--=20 Sent from my Android device with K-9 Mail=2E Please excuse my brevity=2E