Hi Alexander,
5.0.0BETA2 will contain a modification inspired by your patch, which
should actually lower the time during which the lock is held even further.
There would be ways to optimize it much further (I left a todo comment
in the code hinting what needs doing and what the difficulties are), but
decided against doing it now.
Klaus
On 08.04.2015 14:32, [email protected] wrote:
Hello,
I just want to remind about this patch...
Regards,
Alexander
25.02.2015 16:16, [email protected] пишет:
Hello!
We would like to contribute to VirtualBox under MIT license. We want
to suggest our changes for multiple VMs starting speed-up.
The basic idea is to take out getting of host interfaces from write
lock on host object. When many VMs are starting at the same time then
/FindHostNetworkInterfaceByName/ is called for every machine and it
updates network interfaces. So if umount of VMs is large enough then
execution time of this function greatly increases due to lock wait.
There are two /VBoxSVC/ logs in attach: before applying our changes
and after it. In every case we started 80 VMs at the same time.
Statistic of /FindHostNetworkInterfaceByName/ execution time before
changes is:
Min: 0.38 s (nspr-31)
Med: 14.13 s (nspr-82)
Max: 37.70 s (nspr-188)
Avg: 15.07 s
Statistic after changes is:
Min: 1.12 s (nspr-85)
Med: 6.56 s (nspr-228)
Max: 17.27 s (nspr-7)
Avg: 5.82 s
Regards,
Alexander
_______________________________________________
vbox-dev mailing list
[email protected]
https://www.virtualbox.org/mailman/listinfo/vbox-dev
