Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Assess if new machine will meet performance of current machine

1 view
Skip to first unread message

Test Email

unread,
Apr 27, 2012, 8:36:28 AM4/27/12
to
We are planning to migrate our LAMP (more specifically mediawiki) to a
new machine, I would like to test if the new machine is able to handle
the load expected.
I have logs on the current machine, from grouping the logs by every
second, below is some info, if I take each entry in the log as one
connection, at peak I have 92 connections on a given second (the
entire data is : with format num_connections: num_such_seconds {1:
50102, 2: 17593, 3: 7834, 4: 4824, 5: 3487, 6: 2963, 7: 2651, 8: 4045,
9: 4434, 10: 2798, 11: 2024, 12: 1256, 13: 766, 14: 583, 15: 515, 16:
451, 17: 488, 18: 381, 19: 340, 20: 267, 21: 198, 22: 155, 23: 115,
24: 77, 25: 69, 26: 68, 27: 48, 28: 47, 29: 32, 30: 32, 31: 25, 32:
19, 33: 20, 34: 17, 35: 9, 36: 10, 37: 9, 38: 7, 39: 2, 40: 4, 41: 2,
42: 3, 43: 5, 44: 2, 45: 2, 46: 1, 48: 1, 49: 1, 50: 3, 52: 2, 53: 1,
55: 2, 61: 1, 62: 3, 68: 1, 71: 1, 73: 1, 92: 1}
However, if I take number of users making requests by second, I have
the following: with format num_users_connected: num_such_seconds {1:
80318, 2: 22491, 3: 4949, 4: 911, 5: 119, 6: 9, 7: 1}
In the above, The former has connection for css images etc too.
I am using ‘ab’ to load a 314k page, which falls on the larger-size
end of the pages generated (by php dynamically) to test the
performance, however, with this the performance of the machine
degrades when I bump up the simultaneous connections to 6 or above.
Can I know the best way to ensure that the new machine can handle the
load, a brute force way I can think of is to write a script to emulate
the traffic (request by request) I see in the logs (on the current
machine)
0 new messages