Am Dienstag, den 12.06.2007, 07:51 -0400 schrieb Tim Shoppa:

> 36000 requests/min is an average of 600 requests/sec or
> somewhere around 4 or 5kbytes/sec (each request is about 80 bytes).
> 
> If a router cannot handle 4kbytes/sec, should it even
> be considered for the pool? I mean, we try to accomodate
> both large-bandwidth and small-bandwidth connections,
> but 4kbytes/sec is analog dial-up modem throughput.

Hi,

4kbytes/s is not a problem, but 36000 connects/min is.

A lot of routers/firewalls are using linux/iptables. There the default
value of "ip_conntrack_max" is 16k, the default value of
"ip_conntrack_udp_timeout" is 30s. That means the router can handle 16k
new ntp connections in 30s. Every new connection is dropped if the limit
is reached.

If its your own server and your own firewall, you can correct
conntrac_max, but not everyone can configure his companies firewall or
the firmware of his router.

At my server the maximum connections/30s is about 12k when it is in the
global pool(you can see ist with "wc /proc/net/ip_conntrack"). Thats
pretty near the 16k-limit...


Just now my server is in the global or eu pool (6.4k connections,
12kByte/s), so I can have a look at the regional spreading:

600 clients total (the limit of ntpdc's output)
493 TR
 28 DE
 23 US
 11 PT
  6 FR
  5 GB
  3 SE, RU, HU, ES
  2 NL, IT, FI, CZ, CH, CA
  1 ZA, TW, RS, PL, NZ, MY, MK, HR, CN, AT

all turkish Clients are from "ttnet.net.tr"

Max





_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to