On Thu, 16 Nov 2017, W.C.A. Wijngaards wrote:> This release has features for saving memory and faster notification. > With --enable-packed, 33% memory savings could be had, or something > along that size. Notification of secondary servers happens in parallel, > and has faster timeouts. More sockets are used for zone transfers. > This speeds up communication with a larger set of servers.What is the downside of this option? I assume there is one, because otherwise this would not be a compile time option? I'm trying to make the determination for RHEL/CentOS/Fedora on whether I want this compile time option enabled or not. Paul
Hi Paul, On 16/11/17 11:57, Paul Wouters wrote:> On Thu, 16 Nov 2017, W.C.A. Wijngaards wrote: > >> This release has features for saving memory and faster notification. >> With --enable-packed, 33% memory savings could be had, or something >> along that size.? Notification of secondary servers happens in parallel, >> and has faster timeouts.? More sockets are used for zone transfers. >> This speeds up communication with a larger set of servers. > > What is the downside of this option? I assume there is one, because > otherwise this would not be a compile time option?Packed structures are more difficult to access by the compiler. It is likely that the compiler needs to generate more code to access members in the data structure. The assumption that data is aligned in memory is not there. The compiler will generate different code. Because it is packed, it'll use less memory, but that code also has to be executed. I don't know what the speed impact is, if any. Best regards, Wouter> > I'm trying to make the determination for RHEL/CentOS/Fedora on whether > I want this compile time option enabled or not. > > Paul-------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 833 bytes Desc: OpenPGP digital signature URL: <http://lists.nlnetlabs.nl/pipermail/nsd-users/attachments/20171116/8396f038/attachment.bin>
On 2017/11/16 05:57, Paul Wouters wrote:> On Thu, 16 Nov 2017, W.C.A. Wijngaards wrote: > > > This release has features for saving memory and faster notification. > > With --enable-packed, 33% memory savings could be had, or something > > along that size. Notification of secondary servers happens in parallel, > > and has faster timeouts. More sockets are used for zone transfers. > > This speeds up communication with a larger set of servers. > > What is the downside of this option? I assume there is one, because > otherwise this would not be a compile time option? > > I'm trying to make the determination for RHEL/CentOS/Fedora on whether > I want this compile time option enabled or not.If you do enable it, make sure you test on strict-alignment architectures if your OS supports them.