[Haizea] unmatching host CPU information between ONE and Haizea

[Shi Jin]

Here is more information on only one VM can start on a 16 CPU host.

[[code]]
[cloudadmin at frontend-dev var]$ cat ~/templates/centos5.5-64bit.cfg
NAME =centos5.5-x64
CPU = 1
vCPU =  1
MEMORY =1000
...
[[/code]]

haizea.log
[[code]]
[2010-10-14 15:14:23.99] CLOCK   Waking up to manage resources
[2010-10-14 15:14:23.99] CLOCK   Wake-up time recorded as 2010-10-14
15:14:24.00
[2010-10-14 15:14:24.05] RM      Requesting leases
[2010-10-14 15:14:24.05] RM      Running scheduling function
[2010-10-14 15:14:24.05] LSCHED  Used up all future reservations and slot
table is full. Skipping rest of queue.
[2010-10-14 15:14:24.05] CLOCK   Going back to sleep. Waking up at
2010-10-14 15:14:25.00 to see if something interesting has happened by then.


[[/code]]

[[code]]
[cloudadmin at frontend-dev var]$  onehost list
  ID NAME                      RVM   TCPU   FCPU   ACPU    TMEM    FMEM STAT
  16 node5                       1   1600   1592   1592 2467536 2289639   on
  17 node4                       5   1600   1600   1600 6598130 6397532   on
[cloudadmin at frontend-dev var]$ haizea-list-hosts
 ID   Hostname    CPUs    Mem
 1    node5       100     24097.03125
 2    node4       100     64434.8710938
[cloudadmin at frontend-dev var]$ haizea-list-leases
 ID   Type          State      Starting time           Duration      Nodes
 1    Best-effort   Active     Unspecified             36500:00:00:00.00  1

 2    Best-effort   Active     Unspecified             36500:00:00:00.00  1

 3    Best-effort   Queued     Unspecified             36500:00:00:00.00  1

[cloudadmin at frontend-dev var]$ haizea-show-queue
 ID   Type          State      Starting time           Duration      Nodes
 3    Best-effort   Queued     Unspecified             36500:00:00:00.00  1

[[/code]]


Do you know why the host is treated full although there is only one VM
running on it?
Thanks.

Shi
On Thu, Oct 14, 2010 at 1:58 PM, Shi Jin <jinzishuai at gmail.com> wrote:

> It looks like it is implemented here
> lib/python/haizea/core/leases.py:
> 805     def __repr__(self):
>  806         """Returns a string representation of the Capacity"""
>  807         return "  |  ".join("%s: %i" % (res_type,q[0]) for res_type, q
> in self.quantity.items())
>
> only the first instance capacity quantity is output. Shouldn't we output
> the sum of all quantities?
>
> But this should not affect our scheduling, right? I need to look more to
> figure out why haizea has stopped scheduling after the first lease.
>
> Shi
>
>
> On Thu, Oct 14, 2010 at 1:11 PM, Shi Jin <jinzishuai at gmail.com> wrote:
>
>> Hi there,
>>
>> I've set up Haizea-1.0 to work with ONE-1.4. Here is the output
>> cloudadmin at frontend-dev var]$ onehost list;haizea-list-hosts
>>   ID NAME                      RVM   TCPU   FCPU   ACPU    TMEM    FMEM
>> STAT
>>   16 node5                       1   1600   1594   1594 2467536 2256548
>> on
>>   17 node4                       7   1600   1600   1600 6598130 6220621
>> on
>>  ID   Hostname    CPUs    Mem
>>  1    node5       100     24097.03125
>>  2    node4       100     64434.8710938
>>
>> The two hosts each has 16 cores, theirfore OpenNebula report 1600 in
>> TCPU.
>> However, they both show up as 100 in Haizea, which means only 1 CPU,
>> right?
>> Is this a discrepancy? I also find out that I can only deploy a CPU=1 VM
>> on each host, which seems to take the whole resources for Haizea point of
>> view. Is there any thing wrong here?
>>
>> The memory seems to be consistent between ONE and Haizea.
>> Thanks.
>>
>> --
>> Shi Jin, Ph.D.
>>
>>
>
>
> --
> Shi Jin, Ph.D.
>
>


-- 
Shi Jin, Ph.D.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/haizea/attachments/20101014/f4db2419/attachment-0001.htm