dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
151
share rss forum feed

rebus9

join:2002-03-26
Tampa Bay
Reviews:
·Verizon FiOS
·Bright House
reply to serge87

Re: [Northeast] High pings / bad routing...

said by serge87:

@Xtreme2 and serith: I also live outside of Syracuse and am having terrible timing out problems the last week or two. I'd say 1 out of 5 requests to almost any website time out

My trace routes to Google, Craigslist and Reddit(notice the timeouts in hops 3-6):

Long post.... and on page 8 it's late in the game, but I really want to stop and shed some light on traceroutes and how they can be misinterpreted.

Timeouts at intermediate hops mean... pay attention... NOTHING. ICMP is low priority and there are many reasons a router might ignore ICMP if it's busy doing something more meaningful, like crunching BGP table updates. Same goes for high latency at INTERMEDIATE hops.

What matters is DESTINATION results. Is there packet loss to the destination? And is there abnormally high latency? Those are the questions that matter.

Something you cannot see, unless you control both endpoints, is REVERSE PATH. If you ride inbound on Verizon to a particular destination, there is no guarantee the return path will be symmetric-- i.e, that the response will come back to you on the same route. Quite often traffic will take another physical path-- or other carriers altogether.

Why? Traffic engineering at the other end. For example, datacenters often use route selection tools that route based on lowest economic cost, not necessarily lowest latency or best path. Many will balance traffic loads across their circuits. If they see too much traffic outbound on their (for example) Level3 transit, they may shift some of it to (example) Cogent or Telia or whatever carriers they have. OR-- they could route based on lowest cost, meaning they'll send the bulk of their OUTBOUND traffic via (for example) Cogent, while still announcing their prefixes (for inbound traffic) through their premium carriers. So never... NEVER, NEVER... assume to know the reverse path unless you can do a traceroute from the host at the other end.

Now, let's discuss the screenshot you posted. You're near Syracuse so we'll consider that in determining expected latency.

Your trace to google-- best guess on where that server is, is NYC. This one is tough to judge because there are so many possibilities for physical path (the actual fiber route), reverse path, etc. At mid-40 ms latency at hops 4-10, if nothing else it's consistent. Given your physical geography, I would not be entirely surprised if upstate NY Verizon fiber faced Boston (Syracuse -> BOS -> NYC) but without a fiber map it's anybody's guess.

Your trace to craigslit-- note that the servers are in Palo Alto. Assuming your physical path is similar to the google trace, meaning in the vicinity of 45 ms to hit NYC, then onto long-haul fiber across country, latency in the 120 ms range is not unreasonable.

Your trace to reddit-- destination server is in the Washington, D.C., area. Given 45 ms from you to NYC, another 10 ms from NYC->DC is not out of line.

Long post, yes. But so many people mis-interpret traceroutes and/or put too much emphasis on them, I felt it appropriate to pass it through a sanity filter.


nycdave
Premium,MVM
join:1999-11-16
Melville, NY
kudos:14

rebus9,

Great post. Hopefully some people will actually read it and understand what those of us 'in the know' have understood for years....

ICMP is lowest priority, and a standard outbound trace route can never tell the entire story.