* [dpdk-dev] [RFC] ABI breakage for rte_mempool to reduce footprint
@ 2016-01-28 19:02 Wiles, Keith
0 siblings, 0 replies; 2+ messages in thread
From: Wiles, Keith @ 2016-01-28 19:02 UTC (permalink / raw)
To: dev
Hi Everyone,
Currently every mempool created has a footprint of 1.5Megs of memory just for the struct rte_mempool, this also includes all of the rte_pktmbuf creates as well. The issue is the local_cache adds about 1.5Megs of memory, which is a huge amount IMHO for non-cached based mempools. Without the local_cache the structure is about 192bytes. You can set the config option for the cache to ’n’, but then all allocations will not use the per core cache. I have some code I will send as a patch to have the local_cache allocated only when a mempool is created and the caller has a non-zero cache size set in the call.
This will break ABI for the struct rte_mempool, but does remove some of the ifdefs for RTE_MEMPOOL_CACHE_SIZE in the code. The performance appears to be the same, but will do some more testing before I submit the patch.
Please let me know if this would be reasonable or other comments.
Regards,
Keith
^ permalink raw reply [flat|nested] 2+ messages in thread
* [dpdk-dev] [RFC] Abi breakage for rte_mempool to reduce footprint
@ 2016-01-28 20:07 Wiles, Keith
0 siblings, 0 replies; 2+ messages in thread
From: Wiles, Keith @ 2016-01-28 20:07 UTC (permalink / raw)
To: dev
Hi Everyone,
Currently every mempool created has a footprint of 1.5Megs of memory just for the struct rte_mempool, this also includes all of the rte_pktmbuf creates as well. The issue is the local_cache adds about 1.5Megs of memory, which is a huge amount IMHO for non-cached based mempools. Without the local_cache the structure is about 192bytes. You can set the config option for the cache to ’n’, but then all allocations will not use the per core cache. I have some code I will send as a patch to have the local_cache allocated only when a mempool is created and the caller has a non-zero cache size set in the call.
This will break ABI for the struct rte_mempool, but does remove some of the ifdefs for RTE_MEMPOOL_CACHE_SIZE in the code. The performance appears to be the same, but will do some more testing before I submit the patch.
Please let me know if this would be reasonable or other comments.
Regards,
Keith
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2016-01-28 20:08 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-01-28 19:02 [dpdk-dev] [RFC] ABI breakage for rte_mempool to reduce footprint Wiles, Keith
2016-01-28 20:07 [dpdk-dev] [RFC] Abi " Wiles, Keith
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).