Statistics from the NANOG Squid Cache

The table below summarizes the traffic received by the Squid cache at various NANOG meetings:

		HTTP Requests				HTTP Bytes			Median Response Time			Mean Response Time
	#IPs	hits	misses	all	ratio	hits	misses	all	hits	misses	all	hits	misses	all
NANOG19	152	60,014	118,806	178,820	33.6%	376,639,144	961,434,630	1,338,073,774	21	354	231	251	1993	1408
NANOG20	62	29,661	48,511	78,172	37.9%	92,985,036	368,929,707	461,914,743	31	165	104	176	875	610
NANOG21	109	42,232	91,939	134,171	31.5%	325,757,608	1,574,538,418	1,900,296,026	19	193	126	235	1132	850
NANOG22	135	73,023	127,662	200,685	36.4%	354,256,312	1,793,052,883	2,147,309,195	22	250	157	156	1210	826
NANOG23	80	45,306	59,953	105,259	43.0%	223,692,250	903,198,426	1,126,890,676	22	162	103	339	1520	1012
NANOG24	60	42,692	91,529	134,221	31.8%	226,372,793	1,558,537,347	1,784,910,140	15	164	115	106	630	463
NANOG25	87	39,325	71,478	110,803	35.5%	357,321,336	888,902,775	1,246,224,111	27	154	110	213	758	564
NANOG26	93	70,942	124,370	195,312	36.3%	359,444,054	1,597,828,147	1,957,272,201	20	186	125	144	726	514
NANOG27	36	17,986	43,544	61,530	28.2%	81,481,220	590,640,824	672,122,044	8	147	100	364	686	592
NANOG28	10	1,269	5,174	6,443	19.7%	5,952,840	108,016,881	113,969,721	15	156	148	209	2752	2551
NANOG29	97	49,805	99,443	149,248	33%	199,132,266	1,201,302,038	1,400,434,304	10	138	95	157	1,426	1,002

Notes

#IPs is the count of unique IP addresses using the cache. This is a rough measure of the number of users. If the DHCP server recycles IP addresses, and gives the same address to different users, then this count is an underestimate. However, if the server gives the same user different IP addresses, then this may be an overestimate.

Median and mean response times are reported in milliseconds.

NANOG19 was in Albuquerque, and we had four T1's. I think people were more inclined to use the cache given the relativie lack of bandwidth. Note that the response times are higher for this meeting as well.

At NANOG20, we weren't able to make the DHCP server (on Windows) return the WPAD option, so the number of users is significantly less. Also, the DHCP server did not return a domain name, like nanog.host.net, so we couldn't add wpad.nanog.host.net either. People didn't start using the cache until it was announced during the opening session on Monday.

Starting with NANOG23 I've been running a program that measures ICMP RTTs to all hosts in the /22 network block. I have graphs for NANOG23, NANOG24, NANOG25. NANOG26. NANOG27. NANOG28. NANOG29.

For NANOG23 I have Apache logs showing NIMDA probes from two hosts inside our network.