[pve-devel] less cores more iops / speed

[Alexandre DERUMIER]

I can't help you, I'm never done this kind of process profiling :(

Maybe Dietmar can help ?


----- Mail original ----- 

De: "Stefan Priebe" <s.priebe at profihost.ag> 
À: "Alexandre DERUMIER" <aderumier at odiso.com> 
Cc: pve-devel at pve.proxmox.com, "Dietmar Maurer" <dietmar at proxmox.com> 
Envoyé: Mardi 13 Novembre 2012 09:03:26 
Objet: Re: [pve-devel] less cores more iops / speed 

Am 13.11.2012 09:01, schrieb Alexandre DERUMIER: 
> Ok, I hope Intank will look at librbd and find the bottleneck ! 
I hope that too. Sadly i'm not able to produce a callgraph with google 
profiler as the kvm proces just segfaults when it is loaded. 


> ----- Mail original ----- 
> 
> De: "Stefan Priebe" <s.priebe at profihost.ag> 
> À: "Alexandre DERUMIER" <aderumier at odiso.com> 
> Cc: pve-devel at pve.proxmox.com, "Dietmar Maurer" <dietmar at proxmox.com> 
> Envoyé: Mardi 13 Novembre 2012 08:59:41 
> Objet: Re: [pve-devel] less cores more iops / speed 
> 
> Not sure which component limits the speed. There's no measurable 
> component running at 100% CPU load. Or 100% network load... 
> 
> Stefan 
> 
> Am 13.11.2012 08:46, schrieb Alexandre DERUMIER: 
>>> so around 3000 iops by node, I hope It'll improve in time. 
>>>> Yes at least it scales. Sadly only 3000 iops... but each node has a 
>> 
>>>> capacity of 360.000 iops... perhaps a bit overkill... but SSDs are near 
>>>> cheap as SAS drives... 
>> 
>> So, If you launch fio from differents kvm guest on differents kvm host, you are limited by the ceph cluster speed ? 
>> 
>> 
>> ----- Mail original ----- 
>> 
>> De: "Stefan Priebe" <s.priebe at profihost.ag> 
>> À: "Alexandre DERUMIER" <aderumier at odiso.com> 
>> Cc: pve-devel at pve.proxmox.com, "Dietmar Maurer" <dietmar at proxmox.com> 
>> Envoyé: Lundi 12 Novembre 2012 23:37:14 
>> Objet: Re: [pve-devel] less cores more iops / speed 
>> 
>> Am 09.11.2012 15:43, schrieb Alexandre DERUMIER: 
>>>>> To me yes. With 3 Nodes i had a max of 10.000 iops. 
>>> 
>>> Ah, great to known ! 
>>> 
>>> so around 3000 iops by node, I hope It'll improve in time. 
>> Yes at least it scales. Sadly only 3000 iops... but each node has a 
>> capacity of 360.000 iops... perhaps a bit overkill... but SSDs are near 
>> cheap as SAS drives... 
>> 
>>> Did you have tested with gigabit link with bigger latencies ? 
>> No. But i hope i can compare Arista 10G against HP 10G soon. 
>> 
>> Stefan 
>> 
>> 
>>> ----- Mail original ----- 
>>> 
>>> De: "Stefan Priebe - Profihost AG" <s.priebe at profihost.ag> 
>>> À: "Alexandre DERUMIER" <aderumier at odiso.com> 
>>> Cc: pve-devel at pve.proxmox.com, "Dietmar Maurer" <dietmar at proxmox.com> 
>>> Envoyé: Vendredi 9 Novembre 2012 15:28:12 
>>> Objet: Re: [pve-devel] less cores more iops / speed 
>>> 
>>> Am 09.11.2012 11:31, schrieb Alexandre DERUMIER: 
>>>>>> Yes. CPU Load in Ceph for rand. 4k writes can be dropped by 50% when 
>>>>>> disabling all debug setting (see my last post on ceph mailinglist). 
>>>> 
>>>> I just see it, great,I'll test it on my side :) 
>>>> 
>>>> (how many ssds - nodes ?) I hope that Intank is working on random io performance. 
>>> I have 5 nodes with 4 SSDs each. 
>>> 
>>>> I don't understand why we can't get more iops, if they are no cpu or disks bottleneck... (or they are locks somewhere ....) 
>>>> 
>>>> I really would like to known if iops scale with the number of nodes... 
>>> 
>>> To me yes. With 3 Nodes i had a max of 10.000 iops. 
>>> 
>>> Stefan 
>>>