[DRBD-user] DRBD Transport Latency

Ard van Breemen ard at kwaak.net
Tue Jan 30 19:17:47 CET 2007

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On Tue, Jan 30, 2007 at 02:08:07PM +0100, Lars Ellenberg wrote:
> also note that rtt in LAN is usually orders of magnitude
> less than average seek time/fsync time of storage.
> also in a dedicated fo for over 20km,
> rtt might be in the order of one digit ms, iirc.
Gigabit local linux<-switch-2x10Gig fiber 500m-switch->linux (l2 switch = cisco)
64 bytes from 10.41.0.1: icmp_seq=1 ttl=64 time=0.202 ms
64 bytes from 10.41.0.1: icmp_seq=2 ttl=64 time=0.251 ms
64 bytes from 10.41.0.1: icmp_seq=3 ttl=64 time=0.203 ms

Gigabit local linux<->cisco (l2 switch/router = cisco)
64 bytes from 10.53.254.4: icmp_seq=1 ttl=255 time=0.363 ms
64 bytes from 10.53.254.4: icmp_seq=2 ttl=255 time=0.559 ms
64 bytes from 10.53.254.4: icmp_seq=3 ttl=255 time=0.366 ms

Gigabit routed linux<-switch/router-(10Gig fiber>30km)->cisco (l3 router = cisco)
64 bytes from 217.196.40.69: icmp_seq=1 ttl=254 time=0.618 ms
64 bytes from 217.196.40.69: icmp_seq=2 ttl=254 time=0.541 ms
64 bytes from 217.196.40.69: icmp_seq=3 ttl=254 time=0.593 ms

So if you carefully look at the roundtrip times: the distance of
the fiber really doesn't matter that much, only the number of
hops.  A cisco ping roundtrip is longer, since cisco handles
icmp's destined for it at a low priority.



More information about the drbd-user mailing list