>-----Original Message-----
>[Wang, Yipeng]
>Yes, with 2-hash function and 16 entries per bucket, CD has much less misses
>in general.
>
>As first step to combine both CD and DFC, I incorporated the signature and
>way-associative structure from CD into DFC. I just did
>simple prototype without Any performance tuning, preliminary results show good
>improvement over miss ratio and throughput. I will
>post the complete results soon.
>
[Wang, Yipeng] I implemented 8way/bucket with signature each entry. Here is
just some preliminary results (not optimized). There is certain additional
overhead but in general it helps a lot with larger numbers
of flows. With 1 rule/subtable and only 1 subtable, the 1way DFC without
signature always have 100% hit rate thus it is faster, but we assume that is
not the common use case. For other cases, more ways
and a signature per entry should give better performance in general, and
performs more consistent w.r.t. different traffic patterns.
throughput improvement with 8way/bucket and signatures
1rule/subtable
flow/subtable 1 2 3 5 10 20
10k 1.00 0.93 1.18 0.96 0.99 1.00
100k 0.98 1.20 1.45 1.46 1.81 2.41
1M 0.85 1.25 1.32 1.40 1.43 1.09
1.5M 0.76 1.05 1.16 1.27 1.33 1.10
10 rule/subtable
flow/subtable 1 2 3 5 10 20
10k 1.00 0.97 1.00 1.00 1.00 1.00
100k 1.00 0.96 1.08 1.06 1.11 1.16
1M 1.04 1.06 1.09 1.13 1.20 1.27
1.5M 1.04 1.05 1.07 1.09 1.11 1.14
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
