Transcript Document 7682733

Selecting best Redhat Mirror Sites
for parallel download
Jing Yang and Zhong Li
CS522 Semester Project
5/25/2016
Jing Yang , Zhong Li
1
Selecting the best mirror sites
for parallel download
• Investigate traceroute and pathchar to
characterize the top Redhat mirror sites of
interest to UCCS(wait.uccs.edu)
• Apply traceroute to get shortest round trip
time (rrt) values and netowrk topology.
• Apply pathchar to estimate the bandwidth of
the links along which the packet flow.
5/25/2016
Jing Yang , Zhong Li
2
Traceroute and Pathchar Programs
• Traceroute utilizes IP protocol ’time to live’ field
and attempts to elicit an ICMP TIME_EXCEEDED
responses from each gateway along the path to
some host and therefore obtain the round trip
time(rtt). (designed by Jacobson)
• Pathchar correlates rtt with packet length to
estimate the bandwidth. Pathchar provides more
detailed and precise information by sending many
differenet probe packets to each link.
(designed by Van Jacobson)
5/25/2016
Jing Yang , Zhong Li
3
Apply Traceroute for Topology
-70 Redhat mirror websites from north America
were chosen for running traceroute
- ttl is originally set to1 and increased by one each
time the packet is sent to further host.
- ttl is decreased by each gateway and when ttl
becomes zero, the gateways ends back a ICMP
“time exceeded” message.
5/25/2016
Jing Yang , Zhong Li
4
Traceroute Execution Result
-
When we get an ICMP “port unreachable”, does it
mean we have reached the destination?
[zli@wait trace]$ traceroute limestone.uoregon.edu
traceroute to limestone.uoregon.edu (128.223.162.20), 30 hops max, 38 byte
packets
1 cs-content-switch1-router (128.198.60.1) 0.403 ms 1.000 ms 0.833 ms
2 uccsedge (128.198.254.9) 3.153 ms 3.828 ms 2.830 ms
3 uccs-link-at-frgp (128.198.254.2) 3.230 ms 3.361 ms 3.196 ms
4 ucar.edu.ip.att.net (12.124.158.13) 3.752 ms 3.754 ms 3.970 ms
……
17 p10-0-0.mit2.bbnplanet.net (4.1.80.10) 83.610 ms 146.916 ms 83.738 ms
18 b24-rtr-1-backbone.mit.edu (18.168.0.14) 88.859 ms 88.080 ms 88.964 ms
19
radole.lcs.mit.edu (18.201.1.3) 88.724 ms 88.153 ms 91.810 ms
21 rufus.w3.org (18.29.1.67) 268.701 ms 223.507 ms 241.069 ms
5/25/2016
Jing Yang , Zhong Li
5
How Pathchar Works
-
-
local host sends 32 probe packets (by
default) for each different size ranging from
64 bytes to 1500 bytes (45 times)
Some parameters can be passed in the
command line:
-f: initial hop #, -l: max ttl, -q: #queries for more info
check:http://www.cai.org/tools/utilities/others/pathchar/pathcha
rnotes.html
-
It correlates rrt with packet length to estimate
bandwidth
5/25/2016
Jing Yang , Zhong Li
6
Pathchar Execution Example
pathchar to trillian.cc.gatech.edu (130.207.108.135)
can't find path mtu - using 1500 bytes.
doing 32 probes at each of 45 sizes (64 to 1500 by 32)
0 localhost
| 27 Mb/s, 155 us (758 us)
1 cs-content-switch1-router (128.198.60.1)
| 97 Mb/s, 179 us (1.24 ms)
2 uccsedge (128.198.254.9).
…….
14 130.207.251.2 (130.207.251.2)
| ?? b/s, 356 us (94.8 ms)
15 cc-rstone-sys3-156.cc.gatech.edu (130.207.108.158)
| 54 Mb/s, 154 us (95.3 ms), 33% dropped
16 trillian.cc.gatech.edu (130.207.108.135)
16 hops, rtt 93.5 ms (95.3 ms), bottleneck 27 Mb/s, pipe 319241 bytes
~
5/25/2016
Jing Yang , Zhong Li
7
Interpretation of Pathchar Output
- the last line shows the output of this pathchar
run: there are totally 15 gateways involved(16 is
the destination); bottleneck is 27 Mb/s which
occurs between the local host and cs-contentswitch1-router (128.198.60.1)
- Tree diagram shows the result of running
pathchar on all websites(each run takes 4-5
hours)
5/25/2016
Jing Yang , Zhong Li
8
Wait to Redhat Mirror Sites Tree
Wait
26
CS
108
uccsedge
128
uccs
64
141
acr2-sonet2
78
65
acr2-lookc
acr1-lookc
84
iar1-lookc
36
oarnet.cle
oebc2-atm6
208
199.18.115
merit-its
45
atm1-0-0
203
cc-rtr
owen-nero
orstbrdr-gw
55
kedzie-rtr
90
concretel
pa-rtr
17
mirror.pa
28
gbr4-p80s
0.so-3-0-0
ggrl-p370s
55
ubu.nws
186.atm9-0
157
35
Cisco0.xm
19
mirror.pa
sl-bb22-sj
sl-bb20-ana
sl-bb23-ana
1216
sl-bb21-fw
sl-gw40-fw
42
gbr3-p60d
nr1-p360
gr1-p3100
80
35
0.so-2-x11
0.so-3-t11
0.so-2-x12
1237
0.so-3-t12s
148
0.so-6-t11
169
0.so-1-x11
0.so-3-t12l
183
so-1-0-x12
mes1.mae
pacbell
512
34
128.161.3
128
25
n233-150c
ilgw-okc
54
164.58.10
0.os-7-xr1
pos5-0
57
193.atm6
194
191.atm6
67
194.atm8
360
cybertrails
42
cyb-r2-atm0
fillmore
nin.cyber
43
15
c1-pos3-0
gbr3-p60w
c1-pos2-0
333
gbr3-p20n
c1-pos1-n
gbr5-p60n
wbb1-pos2
675
109
130
gbr3-p360n
56
mirror.arc
114
gbr4-p30w
gw2-netrail
cc-rstone
33
ihets-gw
Jing Yang , Zhong Li
n233-150a
2133
3328
atl
15
n233-150n
717
gbr4-p40a
gige2-1
33
208.30
atlas
c1-pos1-c
48
36
29
445
164.58.1
Unit: Mb/s
5/25/2016
gbr2-p100
att-gw
210
35
< 20
20 - 100
100 - 500
500 - 1000
> 1000
gbr4-p80d
275
c6509-core
cis.ohio
5632
1856
256
0.so-0-1-0
c1-pos6-0
331
9984(314)
597
591
gbr4-p70d
296
0.so-0-1-0
120
164.107.126
gbr3-p70
305
53
orstsw1-gw
177
se1-atm3
gbr3-p80
157.at-6-0
0.so-7-0-0
192.205.32
323
405
41
adm-rtr
gbr1-p60
atm2-0
39
corv-car1-g
57
tc4-atm3
xcore2.san
427
484
gbr2-p60
361
57
29
krc5-atm1
acr2-looks
coreouter2
ucar
361
54
trillian
52
12.125.50
1.5
10.252.0
3.3
mirror-no
981
true-love.r
9
Algorithm for Selecting Top Mirror
Sites (Steps 1-3)
1. Sort the all the bottleneck values in
decreasing order and put in array A[total number
of websites]
2. choose the domain name with bottleneck A[0]
as the first website and put in the final result
array S[number of domain names sorted]
3. Find all the nodes where the host with next
bottleneck can merge with other nodes (if
has some). If not put domain name with the
bottleneck in A, choose next value in A and go
back to 3.
5/25/2016
Jing Yang , Zhong Li
10
Algorithm for Selecting Top Mirror
Sites (Steps 4-6)
4. Find all the bandwidths between the merged
nodes and their previous nodes.
5. Compare each bandwidth with the sum of
bottlenecks of hosts that merge, if the sum
does not exceed the bandwidth, put the
domain name with the next bottleneck in
array, choose the next value in array A and go
back to 3.
6. If the sum is larger than bandwidth, choose
the next value in array A and go back to 3.
5/25/2016
Jing Yang , Zhong Li
11
Algorithm for Selecting Top Mirror
Sites (Step 7)
7. Keep doing this until all hosts in array A are put in
array S.
- list of top five websites selected for UCCS
(wait.uccs.edu)
1. true-love.rutgers.edu
2. ubu.nws.orst.edu
3. atlas.services.ou.edu
4. www.cio.ohio-state.edu
5. Fillmore.ind.net
5/25/2016
Jing Yang , Zhong Li
12
Reference
1. http://redhat.com.download/mirror.html
2. http://www.cai.org/tools/utilities/others/pathchar/pa
thcharnotes.html
3. http://wwwiepm.slac.stanford.edu/monitoring/load/pathchar.h
tml
4. http://sitka.triumf.ca/net/pathchar.html
5. http://www.fips.de/Computer/man/pathchar.html
5/25/2016
Jing Yang , Zhong Li
13