|
|
1.1 root 1: /tmp/xxx looks like a 1048576 byte file, but it is all holes except for the last
2: block.
3: :;time cat /tmp/xxx > /dev/null
4: 0.1u 2.4s 3r cat /tmp/xxx
5: :;time cat /n/debug/tmp/xxx > /dev/null
6: 0.2u 20.0s 66r cat /n/debug/tmp/xxx
7: Thus the netfs without disk delay is running at 15887 bytes per second.
8: If the file is buffered in the server there is no noticable change.
9: Thus the slow rate is in the protocol.
10: ws%time cat /n/seki/tmp/__2148 > /dev/null #totally buffered in server
11: 0.1u 21.2s 52r cat /n/seki/tmp/__2148
12: ws%time cat /n/seki/tmp/xxx > /dev/null #hole file, no buffering in server
13: 0.0u 20.8s 53r cat /n/seki/tmp/xxx
14: ws%time cat /n/seki/tmp/xxx > /dev/null #no holes either
15: 0.0u 20.8s 53r cat /n/seki/tmp/xxx
16: the netfs runs at bout 19784 bytes per second, with no perceptible
17: effect of disk or buffering at host end.
18:
19: 1000 local fstats take about 11 seconds of system time,
20: while 1000 getuids take about 2.7 seconds.
21: 1000 remote fstats of /n/debug take 33 seconds of real time, about 10 milliseconds
22: a message.
23:
24: Now if the server and setup are connected through a pipe, instead of the
25: network, catting the file with holes:
26: :;time cat m/tmp/9489 > /dev/null
27: 0.1u 5.1s 12r cat m/tmp/9489
28: without holes:
29: :;:;time cat m/tmp/x > /dev/null
30: 0.1u 5.2s 12r cat m/tmp/x
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.