On 3/7/15 1:22 , Ian Collins via smartos-discuss wrote:
> Ian Collins wrote:
>> Ian Collins via smartos-discuss wrote:
>>> Greg Zartman via smartos-discuss wrote:
>>>> Are there performance issues or hits by sharing a zfs volume from the
>>>> global zone to a kvm instance?  I have several volumes being mounted
>>>> by a centos vm and performance is pretty sluggy. For example, I'm
>>>> getting itermittent loss of connection and pauses in disk read/write.
>>>>     The machine this is running on has fairly respectable hardware, so
>>>> I'm surprised to see this.
>>>>
>>>> Thoughts?   Thanks!
>>>>
>>> Interesting.  I was going to run some tests with this configuration this
>>> weekend.  I was wondering whether using an etherstub for a private LAN
>>> would perform any better than using a IP address on the GZ.
>>>
>>> Might be worth a try.  If you don't, I will
>> Well I did and the numbers don't make for pleasant reading.
>>
>> I tested using bonnie++ in an Ubuntu KVM instance.
> 
> This afternoon I upgraded my test box and ran the test again.  This
> caused the Ubuntu KVM to lock up completely (hogging 100% of a core). 
> Not a good result.
> 
> The VM is still locked up if anyone want's me to investigate why.
> 
> A cursory glance with truss shows cycles similar to:
> 
> /3:    ioctl(16, _ION(0xae, 128, 0), 0x00000000)    = 0
> /3:    ioctl(16, _ION(0xae, 128, 0), 0x00000000)    = 0
> /11:        Received signal #14, SIGALRM, in sigtimedwait() [caught]
> /11:          siginfo: SIGALRM pid=11606 uid=0 code=-3
> /11:    sigtimedwait(0x2010673D0, 0xFFFFFD7FFD7C3C60, 0x00000000) = SIGALRM
> /11:    write(26, "0E\0\0\0\0\0\0\0\0\0\0\0".., 128)    = 128
> /1:    pollsys(0xFFFFFD7FFFDF9150, 9, 0xFFFFFD7FFFDF9330, 0x00000000) = 1
> /1:    read(25, "0E\0\0\0\0\0\0\0\0\0\0\0".., 128)    = 128
> /1:    sigaction(SIGALRM, 0x00000000, 0xFFFFFD7FFFDF92A0) = 0
> /1:    write(9, "01\0\0\0\0\0\0\0", 8)            = 8
> /1:    read(25, 0xFFFFFD7FFFDF92C0, 128)        Err#11 EAGAIN
> /1:    pollsys(0xFFFFFD7FFFDF9150, 9, 0xFFFFFD7FFFDF9330, 0x00000000) = 1
> /1:    read(8, "01\0\0\0\0\0\0\0", 512)        = 8
> 

Most of the KVM threads that we care about are in the kernel. What does
mdb -ke '0t<qemu pid>::pid2proc | ::walk thread | ::findstack -v' show?

Robert


-------------------------------------------
smartos-discuss
Archives: https://www.listbox.com/member/archive/184463/=now
RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb
Powered by Listbox: http://www.listbox.com

Reply via email to