High latencies writing to a memory mapped file

[Matthias Schniedermeyer]

On 27.07.2010 13:59, Kinzel, David wrote:
> >> Matthias Schniedermeyer put forth on 7/27/2010 4:24 AM:
> >> 
> >> > We have a linux-computer with samba to act as a fileserver 
> >for a few 
> >> > Windows-Clients and ever since kernel 2.6.26 (Never 
> >happend with 2.6.25) 
> >> > the server randomly "hangs" for a few seconds (Which is more of a 
> >> > problem that drives people crazy).
> >> 
> 
> The server hangs, or the connections to samba do? I've found that samba

That's a question i actually hadn't though about. This certainly extends 
the circle of possible culprits a little. From the top of my mind thats 
the the NIC (Intel e100) and it's e100-driver and the (As Cheap as 
possible) 24-port 100MBit-Switch used to connect all computers. And last 
but not least the whole TCP/IP network-stack in the kernel.

This also differentiates "our" server to the mentioned "copy" which uses 
whichever GBit-NIC that was onboard (Something from Realtek AFAIR) to 
the a tree of even cheaper 5 and 8 port switches, not to mention 
on-the-fly-wiring.

> is given a ridiculously low IO priority so that any IO on the server
> will pretty much cause it to stall -- be it updating locate, a backup
> job, etc. 

Whatever the culprit is, in our case it's something that changed between 
2.6.25 and 2.6.26, before 2.6.26 we were happy with whatever was current 
for years. (System is running a reguarly updated Debian SID for at least 
6-7 years)

And the most irritating things is the "a reboot fixes it". When the 
problem reared it's head, you can restart samba, drop all caches, 
umount/decompose the RAID and whatnot, the problem immediatly reappears
after you get back to working conditions.




Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.