[00:07] *** eryc has quit IRC (Ping timeout: 255 seconds) [00:20] *** eryc has joined #arpnetworks [00:20] *** eryc has quit IRC (Changing host) [00:20] *** eryc has joined #arpnetworks [02:30] *** novae has joined #arpnetworks [05:34] *** mdu59 has joined #arpnetworks [14:10] how do I interpret the performance numbers on FreeBSD "top" with respect to ARC? [14:11] as in ARC: 15G Total, 2492M MFU, 11G MRU, 1936K Anon, 527M Header, 1485M Other [14:28] ahh - some googling is revealing things [14:28] MFU = most frequently used. MRU = most recently used. [14:31] ... http://www.racktopsystems.com/adaptive-replacement-cache-arc-adapting-to-your-needs/ [14:36] looks like my ghost rates are under 1%. Apparently the RAM-based ARC is doing most of what I need. [14:36] and 91% of my hits are in the MFU cache [14:41] *** sjackso has left [15:02] RandalSchwartz: try sysutils/zfs-stats from ports, for more verbose ZFS stats [15:11] randal: heh [15:12] RandalSchwartz: maybe try the check for sync writes program [15:12] i can't rmeember what it is [15:12] zilstat ? [15:12] http://www.richardelling.com/Home/scripts-and-programs-1/zilstat [15:12] i keep forgetting i can't paste [15:13] why would tmux/weechat show a block cursor when i do middle mouse button, and i have to press enter to get out of it? [15:23] oh, shift middle click fixes it [15:24] it seems i gained "mouse support" [15:31] Yeah - I was looking at zfs-stats [15:31] that's where I got my ghost rates [15:34] zilstat shows sync writes, but looks like it doesn't work on older freebsd at least as dtrace functionality is missing. [15:34] things like mail tend to have sync writes [15:37] vfs.zfs.l2arc_noprefetch [15:38] that's the setting that disables storing streams in l2arc [15:38] ie sequential [15:38] disabling that may help if you're getting low usage. [15:39] vfs.zfs.l2arc_write_max [15:39] vfs.zfs.l2arc_write_boost [15:39] and increasing those [15:39] the first is speed of writing in general, the second is when pool isn't yet full [15:39] err l2arc isn't yet full. [16:23] RMS on "don't use uber": https://stallman.org/uber.html [16:26] drivers can find out who the passenger is? [16:26] That's what she said!! [16:28] yes, Uber tells the driver your first name [16:29] this does ssound scary [16:29] ...how? [16:29] drivers can find out who the passenger is? /articles/2014/03/28/uber-s-biggest-problem-isn-t-surge-pricing-what-if-it-s-sexual-harassment-by-drivers.html [16:29] erk [16:29] how'd that break [16:29] http://www.thedailybeast.com/articles/2014/03/28/uber-s-biggest-problem-isn-t-surge-pricing-what-if-it-s-sexual-harassment-by-drivers.html [16:30] ok it seemss it varies by location maybe [16:30] and you need it in new york, at least. [16:34] hm? [16:35] need what [16:38] full name for some trip report or something [17:08] Yes, they know my first name. [17:08] They often say "Randal?" when picking me up [17:08] makes sense. [17:09] i'm getting tempted to go to 32gigabit network [17:09] But I also know their first name too. And their license plate number and make/model of car. [17:09] i really don't see what benefit there would be :) [17:09] pretty essential to getting picked up [17:09] ebay encourages spending money [17:09] amazon doubly so [17:10] it's like $160 US pre shipping. [17:12] amazon is handy, if they screw up shipping then they tend to refund it. [17:12] or take 20% off the item price. [17:19] yep [17:25] i was reading servethehome before, and apparently someone was selling 18 port 40gigabit ethernet switches for < $1000 [17:26] but most 40 gigabit ethernet cards are like $200 [17:27] err more than that prob [19:10] *** dj_goku_ has joined #arpnetworks [19:10] arrgh [19:10] *** dj_goku has quit IRC (Read error: No route to host) [19:11] I think the Ethernet card in my laptop broke [19:11] [ 29.231153] e1000e 0000:00:19.0 eth0: Error reading PHY register [19:12] it won't send or receive packets anymore [19:12] ethtool says the link isn't detected [19:13] weird [19:14] does it have pxe boot? [19:14] if so enter setup in bios on startup and try changing the timeout value for it [19:14] (to reset eeprom settings) [19:21] wow. that worked [19:21] mercutio: thanks [19:28] yeah not the most obvious thing to do :) [19:40] *** dj_goku_ has quit IRC (Ping timeout: 256 seconds) [19:44] *** dj_goku has joined #arpnetworks [19:44] *** dj_goku has quit IRC (Changing host) [19:44] *** dj_goku has joined #arpnetworks [19:58] which intel PHY is it out of curiosity? [19:58] it looks older than the card i had an issue with :) [19:59] there's a hardware issue w/the 82574L card that can cause it to enter a weird state [19:59] card/module/whatever [20:00] reading intel bug reports on ethernet cards is scary [20:00] then you read the broadcom and they're just as bad. [20:00] my intel ethernet screws up regularly on my motherboard :( [20:22] m0unds: 00:19.0 Ethernet controller: Intel Corporation 82566MM Gigabit Network Connection (rev 03) [20:23] (from lspci) [20:28] that sounds old [20:29] so connecting to a friends server in NY from ARP goes through india: [20:29] http://www.martinbrandenburg.com/2015/paste/2015011002.txt [20:29] that's strange [20:30] looks like issue with AS9498 [20:30] oh [20:30] it's digitalocean's fault :) [20:31] most likely [20:31] yeah it's them [20:31] I told him to tell support about that [20:32] wow 9498 has a lot of routes. [20:33] but yeah it's advertised via airtel in india [20:33] as well as via ntt [20:33] but airtel is over coresite. [20:33] and arp prefers peering to transit routes. [20:33] does HE's LG require javascript... [20:33] no idea [20:34] i'm using arp's looking glass [20:34] apparently [20:34] arp has a lg? [20:34] http://lg.arpnetworks.com/cgi-bin/bgplg?cmd=show+ip+bgp&req=104.131.193.156 [20:34] yeh [20:34] i set it up :) [20:34] return is via NTT [20:34] https://gist.githubusercontent.com/relrod/147cdd932bab34c957c5/raw/2cf34efafb92b76f639a3a76596b1058875ddbe9/gistfile1.txt [20:34] openbsd nice [20:34] yeah from new zealand it's advertised via ntt [20:34] which i checked it against [20:35] we get a few asian routing issues because asia is closer [20:38] hmm [20:39] actually i think it could be more complex than that. [20:39] that's not actually peering i think [20:40] what? [20:40] *** relrod has joined #arpnetworks [20:40] i think it's transit although i have no idea who clear thought is [20:40] but i think r1 was the peering server [20:40] yeah I was wondering that [20:41] not many paths from ARP to India [20:42] even to 8.8.8.8 goes through them [20:42] so it's that it needs to be prepended by digitalocean [20:42] for india at california [20:42] prepended? what's that mean? I don't know much about ARP [20:43] s/ARP/BGP/ [20:43] prepended? what's that mean? I don't know much about BGP [20:43] well quite a few asian providers have peering in california [20:43] you see on the looking glass it shows 3 numbers under aspath ? [20:43] yeah [20:43] if another path has 4 numbers, it'll prefer the path with 3 numbers [20:43] so if you want to deprefer a route you add in an extra number [20:43] okay but that's not necessarily best [20:44] so like 62567 could prepend their asn, so that it showed as 11799 9498 62567 [20:44] err [20:44] so like 62567 could prepend their asn, so that it showed as 11799 9498 62567 62567 [20:44] well the other thing to do is advertise more specific routes, like you can advertise a /24 instead of a /19 [20:44] so they may be using one of the /24s of /19 in india and so advertising there [20:45] or they may just be trying to improve their indian routing. [20:45] if they're trying to improve their indian routing, they can also tell 9498 not to advertise their routes in california [20:45] (but only in asia) [20:45] i dunno who "clear thought" is [20:46] if they're at the front that means ARP peers though right? [20:46] arp gets transit through them it appears [20:46] transit. that's the right word [20:46] err [20:46] creative thought [20:46] http://bgp.he.net/AS11799 [20:47] *** dj_goku has quit IRC (Ping timeout: 264 seconds) [20:47] doesn't help me figure out who they are [20:47] so BGP (I knew this but I wasn't sure how it interacted with routing) doesn't know about pipe size or latency [20:47] http://www.thebgpcollective.com/ [20:47] it seems to be in regard to this [20:47] yeah bgp doesn't know about pipe size or latency [20:48] about the best you can do is use communities. [20:48] basically community is like a tag, so you can say where a route comes from [20:48] or what you want to do with a route [20:49] well do you advertise from your AS at one router in india and at another in new york? [20:49] but if they're different subnets they're different paths to each subnet [20:49] so it's route to subnet not route to AS? [20:49] it's route to subnet yes [20:49] you can advertise your ip addresses from multiple asn's in theory. [20:49] AS is just a tag so you can see who it is? [20:49] yeah. [20:49] it's pretty simple really, that's the issue. [20:50] ok and communities help how? [20:50] because routing can be complicated. [20:50] well like different providers have different ways you can tag your traffic [20:50] http://www.us.ntt.net/support/policy/routing.cfm [20:50] *** dj_goku has joined #arpnetworks [20:50] so like ntt for exam,ple, you can specifiy community of 65403: and it will prepend you in north america [20:51] these are not standardised at all. [20:51] okay so that's like you said DO should tell airtel not to advertise in california. they'd tag with a community? [20:51] well bgp collectgive is a weird thing [20:52] from what i recall, it's to give peering in more locations or such [20:52] but yeah, digitalocean may give a community to not advertise to bgp collective from airtel. [20:53] but airtel would have to set that up to work. [20:53] and publish a list [20:53] airtel have a looking glass :) [20:57] mercutio, can you see other routes in the LG? [20:57] from 9498? [20:57] or from creative thought? [20:57] or in general? [20:57] no like the one through NTT that's longer? [20:57] nah [20:57] it's one of the anonying things atm :) [20:58] actually it shouldn't even be longer. [20:58] That's what she said!! [20:58] digitalocean are directly connected [21:00] http://lg.arpnetworks.com/cgi-bin/bgplg?cmd=show+ip+bgp+as&req=62567 ? [21:01] 2914 is NTT but the 104.131.192.0/19 route isn't through them [21:02] yeah it's a local preference issue [21:02] local preference overrides length [21:02] see how lpref says 150 rather than 100 [21:03] no the IP of where I'm connecting isn't advertised through NTT [21:03] http://lg.arpnetworks.com/cgi-bin/bgplg?cmd=show+ip+bgp+transit-as&req=11799 [21:03] there's a /lot/ of routes from 9498 [21:04] https://lg.airtel.in/ [21:05] there's an informative community of 93 [21:05] i have no idea what 93 means :) [21:06] that thing doesn't seem to work very well [21:06] it's a horrible web site tbut it works for me [21:07] so airtel have peering at nyiix.net [21:07] oh when it says system error it's lying [21:07] and provide transit from india to there. [21:07] yes. [21:07] lieing* [21:07] it's terrible is't it ;) [21:07] lying [21:07] I should learn to spell [21:08] ok so from india it's good route [21:08] but from los angeles it goes via india :) [21:08] so airtel shouldn't advertise routes from nyix in los angeles [21:08] so end result is that airtel is probably doing something wrong [21:09] yes [21:09] and arp is preferring it [21:09] oh I was going on about IP here http://lg.arpnetworks.com/cgi-bin/bgplg?cmd=show+ip+bgp+as&req=62567 not being listed as from NTT. but those last two from NTT are completely different [21:10] so the others _are_ from NTT but I can't see that on lg like you said [21:10] right? [21:10] yeah because the later ones probably aren't advertised to airtel [21:10] *** dj_goku has quit IRC (Ping timeout: 245 seconds) [21:12] *** carvite has quit IRC (Ping timeout: 272 seconds) [21:17] *** carvite has joined #arpnetworks [21:25] *** dj_goku has joined #arpnetworks [21:25] *** dj_goku has quit IRC (Changing host) [21:25] *** dj_goku has joined #arpnetworks [21:31] mkb: i managed to reproduce on a uk vm [21:32] http://pastebin.com/DjApJrSC [21:32] uk to new york should be much less ping than that [21:32] so yeah that may help convince digitalocean that it's their fault :) [21:32] or err airtel's fault. [21:33] relrod, ^^ [21:33] mercutio: I'll update the ticket to include that (the DO VM is mine) [21:33] ok [21:34] it's easy to blame the other end when you don't have other cases :) [21:34] but yeah thanks [21:55] *** novae has quit IRC (Ping timeout: 264 seconds) [21:59] *** dj_goku_ has joined #arpnetworks [21:59] *** dj_goku has quit IRC (Read error: Connection reset by peer)