what's illumos' relationship with XTI/TLI?

are you meaning is it still present and supported? (yes), or do we like it? (not really)

danmcd is probably the one for actual details

Oh...

Yeah... XTI/TLI still exist.

Biggest mission-critical user is NFS & friends AFAICT. TCP/IP itself is direct-sockets these days, unless you have REALLY OLD CODE that either uses TLI/XTI or worse, does socket() followed by an I_POP of "sockmod" which means you were optimizing for pre-Solaris-2.6.

We highly recommend you NOT use XTI/TLI.

@Molnija that answer your question?

very thoroughly, thanks.

So, except for a toy, don't write new code that uses that interface?

YW. And yes, your last sentence is correct.

Understood, understood.

So were XTI and TLI kinda like a collective delusion or something? That's the energy I get from reading about it.

damn, my music player skips a lot. I should get a pair of SSDs, or a rank of SAS rotaries

Molnija... I wasn't involved in Solaris or SysV when that decision was made. The idea was AT&T's SysV folks REALLY LIKED STREAMS (when you were shouting STREAMS), and thought it would take over the world.

hehe

They included "sockmod" and "libsocket" for compatiblity with BSD sockets apps.

Now in a world with not many connections and data dominance, it wasn't HORRIBLE.

... and popping sockmod was a way of stripping the BSD compatibility after you'd used it?

Popping sockmod was done after acceptance/connection because read/write didn't need any of its functionality.

ah

Early web deployments shined a light on how horrible it was. Apps were used to the sockets API, and a TCP accept() employed 6 (yes six) syscalls under sockmod/libsocket.

woof!

Sounds like something I should be glad for the death of

In Solaris 2.6 sockfs was developed, and while it kept the STREAMS TCP/IP kernel code, it obviated libsocket and sockmod.

(For historical purposes... SysV dropped in the late 80s, Solaris 2.0/SunOS 5.0 was 1991, and Solaris 2.6 was 1997.)

After sockfs bought us some time, the FireEngine project, while not perfect, was a good de-streams-ing of the TCP/IP stack. The Nemo (GLDv3) project finished the device-driver side of the job. And the Volo (direct-sockets) project finished the API/sockfs side of the job.

aha

And finally, the IP datapath refactoring (concurrent with, but required for Volo) cleaned up TCP/IP's paths even more.

IP datapath and Volo were the last Big things to come out of Sun before Oracle closed things. Nemo, BTW, introduced vnics as well.

right on

Thanks for the talk. I think I understand the situation a little better. Sorry for wasting your time on this. >.<

You did not waste my time (esp. two days before the US Thanksgiving holiday), I hope I didn't waste yours (and I REALLY hope I didn't get any details wrong).

(And pardon the latency Molnija...)

you have not wasted my time at all

it was interesting to read

maybe one day alanc &co will write us an official history

a

dan was much closer to all that than I was

I just have the benefit of still having SCCS history to look at

Yes, I meant of everything.

he was part of the &co

"In the beginning there was BSD, and wnj said it was good..."

exciting?

wnj?

Bill Joy?

yes

ah

richlowe: not really, just time to remove libXevie and libXxf86misc since X.Org abandoned them several years back

(My "30 years in this business" anniversary is 14mos away.) Gotta run, errand.

you've been ITB longer than I've been alive

I guess iI can not say "in this business" as I did start working in university:D

now danmcd feels really old :D

I'm just a smol dragon

back then, we got 2x sun3 workstations.

but my first unix account was created in SCO unix...

fun time it was.

wew.

my first account was on DEC Ultrix, my first root was on Apollo Domain/OS

my first OS-level source commit was (big surprise) Solaris

does our nfs server support proto=rdma?

looks like we dont