About ::
TODO ::
Blog ::
RSS ::
Old blog ::
Projects ::
GIT ::
Gallery ::
Notes
Wed, 11 Oct 2006
Climbed a little in "Extreme".
It was quite easy training, since the climbing zone was occupied by climbing school,
so I only climbed some traverses on "Kaskad" sector. Although they all have
quite complex rate, I think they are simple traces from technical point of view,
and require only a lot of power in the most cases. So I wipped my fingers off about three hours
and completed three or four traverses and couple of boulderings.
I prefer more technical traces actually, so it was not that perfect training, but
nevertheless it was a good time there.
Sending support for userspace netchannels.
Here is first tcpdump output for
userspace network stack combined with
netchannels.
16:44:39.584216 IP (tos 0x10, ttl 64, id 32988, offset 0, flags [DF], proto: TCP (6), length: 60) 192.168.4.78.40394 > 192.168.0.48.smtp: S, cksum 0xbe4d (correct), 1757948353:1757948353(0) win 4096 [timestamp 1160570888 0,nop,nop,wscale 8,nop,mss 1460] 16:44:39.584690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto: TCP (6), length: 60) 192.168.0.48.smtp > 192.168.4.78.40394: S, cksum 0xd745 (correct), 370677133:370677133(0) ack 1757948354 win 5792 [mss 1460,nop,nop,timestamp 281558316 1160570888,nop,wscale 7] 16:44:43.785198 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto: TCP (6), length: 60) 192.168.0.48.smtp > 192.168.4.78.40394: S, cksum 0xd32b (correct), 370677133:370677133(0) ack 1757948354 win 5792 [mss 1460,nop,nop,timestamp 281559366 1160570888,nop,wscale 7]Userspace netchannel stack is trying to setup a connection with Linux test machine.
Note that netchannel does not read those ACKs sent by Linux host, since I have not completed file descriptor allocation for netchannels, so reading works only after connection has established and it is possible to bind new file descriptor to netchannel. I will resolve this issue tomorrow and will start to test alternative TCP/IP stack in userspace with netchannels.
/devel/networking :: Link / Comments ()
New kevent benchmark from Johann Borck.
kevent
web server behaves slightly better than epoll
with new kevent based client application.
Both scale extremely good,
but there is still some performance decrease, probably related to memory footprint
and/or tree traversal speed, although kevent stays on the same line
(it has smaller memory allocations and size).
/devel/kevent :: Link / Comments ()
My hammock has arrived!
To set it up I only need to find couple of anchoring bolts
and connection chain, I hope I will get them tomorrow in the local
development shop so it will be possible to test my new bed.