OVH Community, your new community space.

RPS: QoS & co


Andy
13-04-2008, 16:09
Hello,
We have done different tests on 192/195 and 196/199.
On the 192/195, we have no problem now.
Everything works perfectly. 196/199 on hand, we
Have a few clients who come by to destabilize
When the server iSCSI. This is due to the use of a
Server type (soft, we have not tested in beta).

We have done tests on the QoS so that the limit
Kind of use, but the protocol detects the iSCSI
QoS management in the sense 'DROP'. Basically, it will destabilise.
It is not very serious since it is trying to destabilize one
Bourinne client that the server iSCSI. But the fact of
Destabilize causes an overload of the server that iSCSI
Bump because of destabilisation more than usual.

We are going to change the QoS of 'DROP' to 'WAIT'
And delay packets of those who bourinnent. We should
Able to finalize the testing on a test platform
In a few days.

However, this morning I wondered if it is not
Simpler and clearer for all to move the interface
RPS network of 10Mbps to 100Mbps. Thus we have the 'WAIT'
Natural and perfect, the offer is clear on the site and it
No gender souentendu RPS 100Mbps and 10Mbps on the iSCSI
Or QoS or I do not know. What do you think?

Yours
Octave

oles@ovh.net
12-04-2008, 09:48
Hello,

We have done different tests on 192/195 and 196/199. On the 192/195, we have no problem at the moment. Everything works perfectly. However regarding 196/199, we have a few clients who succeed to destabilize sometimes the iSCSI server. This is due to the use of a type of server (an application we have not tested in beta).

We have done some tests on the QoS in order to limit this kind of use, but the iSCSI protocol detects the QoS management as 'DROP'. Basically, it destabilises it. It is not very serious since we are trying to destabilize a customer that slows the iSCSI server. But the destabilization overloads the iSCSI server, which works due to a greater than usual destabilisation.

We are going to change the QoS of 'DROP' to 'WAIT' and delay the packets of those that slow down. We should be able to finalize the testing on a test platform in a few days.

However, this morning I wondered if it is not simpler and clearer for all to move the RPS network interface from100Mbps to 10Mbps. Thus we have the natural and perfect 'WAIT', the offer is clear on the site and there is no undercurrent such as 100Mbps on the RPS and 10Mbps on the iSCSI Or QoS or what have you. What do you think?

Regards,

Octave