Squid 2.4-DEVEL-ntlm performing MUCH worse than 2.3-STABLE?

While deploying squid-2.4-DEVEL-ntlm, I noticed that it seemed to perform
MUCH worse
than 2.3-STABLE1 (I'm talking about a factor of 50 - roughly the same HTTP
load
costs with 2.4 50 times the CPU it does with 2.3)
It appears to be true, however I have NO clue as to why it happens.
In particular, it _seems_ (I might be misinterpreting the profiling output)
to be spending inordinate amounts of CPU time in comm_poll:

Flat profile:
  % cumulative self self total
 time seconds seconds calls us/call us/call name
 61.87 5.81 5.81 1131303 5.14 7.06 comm_poll
  7.67 6.53 0.72 8426632 0.09 0.16 commDeferRead
  3.51 6.86 0.33 1131303 0.29 0.29 storeDirCallback
  3.41 7.18 0.32 4285511 0.07 0.07 clientReadDefer
  2.02 7.37 0.19 1131303 0.17 0.17 eventNextTime
  1.38 7.50 0.13 1131303 0.11 0.82 eventRun
  1.28 7.62 0.12 main
...

Call graph:
index % time self children called name
...
                5.81 2.18 1131303/1131303 main [1]
[2] 85.1 5.81 2.18 1131303 comm_poll [2]
                0.72 0.61 8421295/8426632 commDeferRead [3]
                0.33 0.00 1131303/1131303 storeDirCallback [7]
                0.08 0.10 1131302/1256566 statHistCount [12]
                0.00 0.10 780/780 commHandleWrite [22]
                0.00 0.10 513/513 httpReadReply [23]
                0.01 0.06 108594/108594 comm_poll_dns_incoming [30]
                0.00 0.04 295/1994 clientReadRequest <cycle 1>
[41]
                0.01 0.00 66647/66647 comm_poll_icp_incoming
[102]
                0.00 0.01 75/1994 commConnectHandle <cycle 1>
[156]
                0.01 0.00 1114/1819 ignoreErrno [111]
                0.00 0.00 5337/5337 comm_poll_http_incoming
[155]
                0.00 0.00 11/11 pconnRead [182]
                0.00 0.00 93/93 checkTimeouts [198]
                0.00 0.00 28/28 snmpHandleUdp [233]
                0.00 0.00 2/2 helperHandleRead [263]
                0.00 0.00 2026/2026 fdIsIcp [534]
                0.00 0.00 2026/2026 fdIsDns [533]
                0.00 0.00 1749/1749 fdIsHttp [540]
...

However, a quick comparision on that function didn't show up any significant
change since 2.3STABLE.

Any idea on what's going on?
Thanks in advance.

        Francesco Chemolli
Received on Mon Jul 24 2000 - 04:09:00 MDT