SixXS::Sunset 2017-06-06

Ticket ID: SIXXS #761028
Ticket Status: User

PoP: fihel01 - DNA Oy (Helsinki)

High latency peaks
[fi] Shadow Hawkins on Tuesday, 08 July 2008 17:45:08
I've seen some ssh-hurting latency peaks with fihel01 PoP. My handle is JK22-SIXXS, the tunnel is T13848. My router is Debian with Linux 2.6.25.10 vanilla, end host is Mac OS X (gigabit ethernet LAN). My Internet connection is 24M/2M ADSL2+. Hopefully code in this system means fixed-width font pre-formatted text. I'm pasting mtr traceroutes inside a code-block. Please monitor fihel01 latency (short-term, not average). The latency peaks are short but annoying.
IPv6: Last 65 pings 1. cl-515.hel-01.fi.sixxs.n ................................................................. 2. gw-515.hel-01.fi.sixxs.n ????.?????????????????????????????????????????????????b?????????? 3. van5r5.ip6.fv.fi .................aaa.b...c..............a...a.......1.b.b.a.b.... 4. hki1r5.ip6.fv.fi .................222.b.2.b..............3...3.........a.b.1.a... 5. 2001:7f8:7:1::6667:1 .....................a...a....1.........1.........1...3.a...a... 6. ge1-3-0-997.pe1.pa1.hki. 1........2........2..a...a.1........333.b...b3........3.a..>3..2 7. sr1.alto.esp.fi.v6.eunet ......2..........1.3.1...2....2.........a2..a......333.>...>.... 8. xob.kapsi.fi ........................................a...a....33....>.b.c.... Scale: .:22 ms 1:56 ms 2:89 ms 3:168 ms a:370 ms b:560 ms c:896 ms IPv4: Last 65 pings 1. 172.20.100.2 ................................................................. 2. dsl-olubrasgw1-fe40fb00- 11111111111111111111111111111111121111111111132111111111111111111 3. 141.208.151.153 1>11111111111111111111111111111111111111111111111111111111111acc 4. olucore1-e-0-1-0.datanet 1111111211111111111211111111111211111111111111111111111111111111 5. hkicore1-e-1-1-0.datanet 333333333333333333333333a33333333333333333333333333a333333333333 6. hkiasbr1-s-0-0-0.datanet 3b33333>3333a?3333333?a33?333?3333b3333a3333a3333a?333a333?33c?a 7. elisa.ficix1-ge.ficix.fi 33333333333333333a3333333333333333333333333333333333333333333333 8. ge1-3-0-10.sr1.alto.esp. 33c3333333333a333333333333b3333333333c3333333333a333333333333c33 9. xob.kapsi.fi 3333333333333333333333333333333333333333333333333333333333333333 Scale: .:4 ms 1:9 ms 2:15 ms 3:29 ms a:64 ms b:97 ms c:156 ms mtr to PoP: Last 65 pings 1. 172.20.100.2 3....1.................................................. 2. dsl-olubrasgw1-fe40fb00- 111113111111111111231111111a1113112111111111112111111111 3. 141.208.151.153 112c3a31111111111111111111111111111111111111111111111111 4. olucore1-e-0-1-0.datanet 11311111111111111111111111111111111111111111112111111111 5. hkicore1-e-1-1-0.datanet 323333322322a2323333333232323233332233232a32323323222332 6. hkiasbr1-s-0-0-0.datanet 333333222322223332232223a2222a3b32223323ac2?32a323322222 7. dna.ficix1-ge.ficix.fi 33333233233232222222332333233323232333232232232233332323 8. van5r6.k.fv.fi 333333333333333333333333333333333a3333333333333333333333 9. fihel01.sixxs.net 33333333333>>>33333333b333a33333333333333333333333333333 Scale: .:4 ms 1:10 ms 2:16 ms 3:30 ms a:66 ms b:100 ms c:161 ms
State change: user Locked
[ch] Jeroen Massar SixXS Staff on Tuesday, 08 July 2008 17:48:55
Message is Locked
The state of this ticket has been changed to user
High latency peaks
[ch] Jeroen Massar SixXS Staff on Tuesday, 08 July 2008 17:55:54
Those latency peeks can come from a *lot* of places (eg your DSL modems network queue), on every hop between you and the PoP and as you are not connecting to the PoP, also to the hosts behind that, which gives another couple of hops where this can be happening. It thus can be just about anywhere. Now if we had copious amounts of free time we could maybe keep an eye on it; unfortunately we don't. According to your mtr the first IPv6 hop is already 22ms, that is much slower than most DSL links already. The PoP has question marks all over the place which is odd too. The high ping responses all outside of the PoP and thus could happen anywhere. Note that pings are two-ways: sent and received, it is quite hard to determine if this is happening outbound or inbound.
High latency peaks
[fi] Shadow Hawkins on Tuesday, 08 July 2008 18:04:57
Please re-evaluate the status. I believe this affects all fihel01-users. I can't help living in Northern Finland and you only having a PoP in South .fi. 12ms punishment for that, 5ms for DSL. I have ~17ms latency to fihel01.sixxs.net. From the mtr to PoP I think it can be reasonably said that the latency seems to occur at the last hop (middle routers are not very reliable indicators unless multiple in row since they have low priority for ICMP). The last hop (fihel01) has multiple > values. Question marks are likely ICMP rate-limiting and can be ignored when they occur only on one host. I believe this latency occurs on fihel01. This is not limited to myself only. I also have mtr from a friend's host using SixXS tunnel. He has a completely different ISP over IPv4.
Last 65 pings 1. 2001:14b8:100:66::1 ????????????????????????????????????????????????????????????????? 2. 2001:14b8:0:3007::2 .....21a.b...b..aaa..b...b..221.b..ac..3.......3..2a..a...a..3... 3. 2001:14b8::4 .....1.3.b...b..3a3.>a..>b......b.33b..1a......1...3..3...31.1... 4. 2001:7f8:7:1::6667:1 .......1.a...a..3...>a..>a....1.a11.b...a.............1...1...... 5. ge1-3-0-997.pe1.pa1.hki. 2.b..1c..a...a......>3..>a......a...a...3......3..........1...... 6. 2001:670:3:7:195:197:59: ..a...c..3...3......c1..c22.....3...a..11.........b.....a....... 7. 2001:670:a0:17::2 ..3.aab.c...c...332.c...c...........3.a.....3.a...b.....3b1.a... Scale: .:44 ms 1:90 ms 2:136 ms 3:244 ms a:522 ms b:784 ms c:1247 ms
High latency peaks
[fi] Shadow Hawkins on Tuesday, 08 July 2008 18:25:50
To make it even more clear: here's a screenshot with two mtrs running at the same time on two different hosts behind two different ISPs. The connecting factor is fihel01. There's a relatively strong correlation of latency peaks. I'm not saying it's 100% but it's enough to suspect that it's a problem of fihel01, not either one of us. You have my word as well that this correlation is also true for other time periods than that short sample. Two mtrs screenshot (Please ignore the 1s and only concentrate on high latency values that are close in time, that is vertical lines of latency).

Please note Posting is only allowed when you are logged in.

Static Sunset Edition of SixXS
©2001-2017 SixXS - IPv6 Deployment & Tunnel Broker