On Mar 16, 2015, at 5:44 AM, Sagi Grimberg wrote:
> On 3/13/2015 11:28 PM, Chuck Lever wrote:
>> Acquiring 64 MRs in rpcrdma_buffer_get() while holding the buffer
>> pool lock is expensive, and unnecessary because most modern adapters
>> can transfer 100s of KBs of payload using just a single MR
On 3/13/2015 11:28 PM, Chuck Lever wrote:
Acquiring 64 MRs in rpcrdma_buffer_get() while holding the buffer
pool lock is expensive, and unnecessary because most modern adapters
can transfer 100s of KBs of payload using just a single MR.
Instead, acquire MRs one-at-a-time as chunks are registered
Acquiring 64 MRs in rpcrdma_buffer_get() while holding the buffer
pool lock is expensive, and unnecessary because most modern adapters
can transfer 100s of KBs of payload using just a single MR.
Instead, acquire MRs one-at-a-time as chunks are registered, and
return them to rb_mws immediately duri