Re: Master is still *really* slow:-/
Jason Gunthorpe <jgg@gpu.srv.ualberta.ca> writes:
> On 4 Nov 1998, Philippe Troin wrote:
>
> > Well, I don't go through sprintlink at all *and* I have a lot of
> > packet loss...
> >
> > For comparison:
> > Packets Pings
> > Hostname %Loss Rcv Snt Best Avg Worst
> > 1. tantale.fifi.org 0% 101 101 0 1 4
> > 2. sjw-pm3-01.blueneptune.com 0% 101 101 145 155 231
> > 3. sjw-t1-01.blueneptune.com 0% 101 101 131 154 198
> > 4. main-106.sjc.above.net 0% 101 101 143 199 798
> > 5. core2-main.sjc.above.net 0% 100 101 140 162 387
> > 6. savvis-above-45Mbps.sjc.above.net 0% 100 100 140 158 278
> > 7. atm9-0-031.CR-1.DllsTX.savvis.net 6% 94 100 186 219 603
> > 8. IGLOBAL-1.DllsTX.savvis.net 8% 92 100 187 195 244
> > 9. master.debian.org 7% 93 100 189 201 407
> >
> > IMHO, savvis is the culprit. The problem started when master started
> > using savvis too...
>
> Savvis has this fun way of asymetrically routing. Try mtr from master to
> you, it will be different. This IMHO is one of the leading causes of
> troubles. We can't do anything without some hard reproducable evidence
> then we can bop the savvis people and they will fix it.
Yeah, you got it... The packets on thei way back go through
sprintlink... and experience a bit of packet loss:
Hostname %Loss Rcv Snt Best Avg Worst
1. outrout.novare.net 0% 109 109 0 1 5
2. GA0FF.GR-1.DllsTX.savvis.net 6% 103 109 1 24 512
3. sl-gw13-fw-10-0-T3.sprintlink.net 5% 104 109 4 10 329
4. sl-bb11-fw-2-2.sprintlink.net 6% 103 109 2 5 8
5. sl-bb22-ana-5-0.sprintlink.net 8% 101 109 26 29 69
6. sl-bb21-ana-9-0.sprintlink.net 8% 101 109 26 29 69
7. sl-bb10-sj-0-0.sprintlink.net 8% 101 109 34 37 69
8. sl-gw2-sj-0-0-0-155M.sprintlink.net 9% 100 109 35 41 177
9. sl-abovenet-3-0.sprintlink.net 11% 98 109 53 64 216
10. main2-core2.sjc.above.net 16% 92 109 54 61 92
11. gw.blueneptune.above.net 13% 94 108 54 62 92
12. sjw-pm3-01.blueneptune.com 17% 90 108 57 64 78
13. ppp-gw.fifi.org 17% 90 108 209 259 591
14. ceramic.fifi.org 13% 92 108 243 263 595
Now that's real assymetric routing :-)
Looks like something's broken to me (sprint ?).
Phil.
Reply to: