181 reputation
110
bio website luca.pierobon.name
location Italy
age
visits member for 2 years, 2 months
seen Jul 7 at 9:50

Jul
3
awarded  Popular Question
Mar
24
comment How can I get CentOS 5 to recognize an increase in drive space on my VM?
Just another note: if your partition is referred by label, /etc/fstab could not be the best way to check for LVM usage, that's why I usually double-check it with pvdisplay/lvdisplay
Mar
24
comment How can I get CentOS 5 to recognize an increase in drive space on my VM?
To confirm you are running LVM you can also check the output of pvdisplay (to see if there are Physical Volumes defined) or lvdisplay (to see if there are Logical Volumes defined). You can also look at /etc/fstab to check what kind of partitions are referred to: if you only have real partitions (/dev/hd*, /dev/sd* for instance) you are probably not using LVM; otherwise you'll see references to some LVM logical volume (for instance /dev/GroupName/VolumeName)
Jun
6
revised Does CGi / FastCGI improve when I increase MinSpareServers?
Added more info taken from the official Apache2 documentation dealing with the directives we are talking about
Jun
5
answered Windows 7 Samba issue
Jun
5
answered Server Load very high despite CPU usage being low
Jun
5
revised Does CGi / FastCGI improve when I increase MinSpareServers?
reworked the conclusion of my reply
Jun
5
awarded  Commentator
Jun
5
comment Does CGi / FastCGI improve when I increase MinSpareServers?
of course, most of the reasoning is related to what is being served dynamically by apache, which is often the biggest part.. Static files serving is a simpler task, as compared to that, so I didn't take it in account
Jun
5
answered Does CGi / FastCGI improve when I increase MinSpareServers?
Jun
5
answered Rebuilding Raid-arrays
May
6
comment Too many established connections left open
My FW admin said that, besides authorization rules (which of course are met here, since the connections get established), the only rule is a 1-hour timeout on TCP connections... so the problem must be on application side, which is not closing connections when done. From what I can tell they stay open till they get dropped by the firewall, and after that nothing can bring them down, other than closing the application's socket server-side (killing it..). I still haven't solved, but your assistance was valuable to direct me to the right way! I'll let you know how it ends :)
May
6
awarded  Scholar
May
6
accepted Too many established connections left open
May
4
comment Too many established connections left open
I think we're getting closer: let's see what my firewall admin will say.. I'll give you more feedback as soon as I get some reply. Thanks in the meanwhile!
May
2
comment Too many established connections left open
As suggested by @Zabuzzman, I've observed with tcpdump a couple of connections: the former exchanges data about every 2 minutes, so it is effectively active; the latter exchanged data once and then was silent for at least half-an-hour.. until I stopped monitoring it. Both were still listed as "ESTABLISHED off" within netstat..
May
2
comment Too many established connections left open
Assuming your hypothesis is right, and given I'm still not much into firewall's internal mechanics, is it correct that a firewall drops connection without closing them as the TCP protocol expects? I can agree on closing too-long-inactive connections, but I thought that before purging a connection from the firewall tables it should be given some kind of "closing" signal..
May
1
comment Too many established connections left open
Ok, this is a good starting point. So if connections get cut they won't be released, unless a) the server's service is restarted or b) the connections are created with keepalive, right? Is there a way to force this app using keepalive connections instead of "plain" ones?
May
1
comment Too many established connections left open
Maybe this can be relevant to the discussion: the system is behind a corporate firewall and inside a NAT..
May
1
comment Too many established connections left open
So are you telling me that when a TCP connection is in the "established" state and isn't using keepalive, it won't be released by the server if the client goes silently away, for instance by means of a firewall which "cuts the line"? Shouldn't it be closed when the firewall closes it? Besides that I thought there was some sort of timeout in the server's TCP stack which tells how long can a connection stay there without activity... By the way there is a corporate firewall between this server and their clients..