DRBD sync speed.
Ivan Shevelenko
Ganeti does have a parameter to setup drbd sync speed?
I Have a 1G interface, but sync rate:
speed: 95,136 (61,920) K/sec
Iustin Pop
That is pretty good, actually. 1G interface has a maximum speed of
~125MB/s, which is 122,070 K/sec.
Ganeti limits each DRBD connection to 60,000 K/sec, so as to not
overload the network. If you are sure you want to change it, then you
can modify the constants.py file. But unless you have a 10G network, I
don't think you'll gain too much (what IO system do you have)?
regards,
iustin
Ivan Shevelenko
>>
>> I Have a 1G interface, but sync rate:
>>
>> speed: 95,136 (61,920) K/sec
>
> That is pretty good, actually. 1G interface has a maximum speed of
> ~125MB/s, which is 122,070 K/sec.
>
> Ganeti limits each DRBD connection to 60,000 K/sec, so as to not
> overload the network. If you are sure you want to change it, then you
> can modify the constants.py file. But unless you have a 10G network, I
> don't think you'll gain too much (what IO system do you have)?
I understand. Thanx.
ps. I have a dell perc raid 0 on sata2 disks.
Lance Albertson
Have you considered adding the ability to change that via gnt-cluster
modify? It might be beneficial to folks who have high enough
bandwidth/IO. I noticed this early on as well. Its not a big issue
though :)
--
Lance Albertson
Systems Administrator / Architect Open Source Lab
Network Services Oregon State University
Guido Trotter
> On 12/03/2010 08:28 AM, Iustin Pop wrote:
>> On Fri, Dec 03, 2010 at 08:03:28AM -0800, Ivan Shevelenko wrote:
>>> Hi all.
>>>
>>> Ganeti does have a parameter to setup drbd sync speed?
>>>
>>> I Have a 1G interface, but sync rate:
>>>
>>> speed: 95,136 (61,920) K/sec
>>
>> That is pretty good, actually. 1G interface has a maximum speed of
>> ~125MB/s, which is 122,070 K/sec.
>>
>> Ganeti limits each DRBD connection to 60,000 K/sec, so as to not
>> overload the network. If you are sure you want to change it, then you
>> can modify the constants.py file. But unless you have a 10G network, I
>> don't think you'll gain too much (what IO system do you have)?
>
> Have you considered adding the ability to change that via gnt-cluster
> modify? It might be beneficial to folks who have high enough
> bandwidth/IO. I noticed this early on as well. Its not a big issue
> though :)
>
There is a design for "disk parameters" but unfortunately we never got
around to implement it, yet.
So yes, we considered it, but for now haven't written the code. :) Of
course an implementation would gladly be accepted.
(we can even think of having the parameters initially only at cluster
level, to simplify things, let's say for 2.4, and then to move them to
per-instance if/when needed later on)
Guido