[ovs-dev] jsonrpc and unbounded amount of cpu

Michael C Cambria mcc at fid4.com
Mon Mar 16 16:23:20 UTC 2020


On 3/13/20 2:03 PM, Ben Pfaff wrote:
> On Tue, Mar 10, 2020 at 03:42:30PM -0400, Michael C Cambria via dev wrote:
>> The input_buffer in struct jsonrpc was recently increased to 4k [ https://github.com/openvswitch/ovs/commit/ea5c1ba0e3b899b8b6684f23a44bbfd4331815ee
>> ]
>>
>> Years ago there was a change [ https://github.com/openvswitch/ovs/commit/3a4548cfb2fb7e7344618f449ab9bf651ef83e6b
>> ] to limit how much cpu jsonrpc_recv() could take before yielding back to
>> its caller.  Should the value of 50 be adjusted down to compensate or is
>> this no longer a problem?
> This is a reasonable question (although I'm curious how it came to
> mind).


Code inspection, looking for bottlenecks.


> Commit 3a4548cfb2fb ("jsonrpc: Keep jsonrpc_recv() from taking over 
> the CPU.") changed jsonrpc_recv() from using a potentially infinite 
> amount of CPU to a finite amount. The bound that it chose was 
> arbitrary (25 kB). Multiplying it by 8 (200 kB) is also arbitrary. I 
> don't have a reason to believe that either figure is better than the 
> other.


More information about the dev mailing list