[ofa-general] Performance of UDAPL RDMA vs IB verbs
Arlin Davis
ardavis at ichips.intel.com
Thu Feb 14 16:45:55 PST 2008
Arlin Davis wrote:
> Chuck Hartley wrote:
>> We are doing performance measurements on an application that is using
>> uDAPL RDMA reads for some large transfers and the BW is less than we
>> expected. The transfers are 4MB and we are seeing BW of 930MiB/sec
>> (DDR). When we do the same transfer size using ib_read_bw we get 1475
>> MB/sec. On a pair of machines with SDR interfaces, we get 697MiB/sec
>> and 918MB/sec respectively.
Here is a quick comparison of verbs, rdma_cm, and uDAPL using
ib_read_bw, ib_write_bw, ib_rdma_bw -c, and dapltest. My
results are very close using default size of 65536.
1.) IB verbs: ib_read_bw, ib_write_bw
------------------------------------------------------------------
RDMA_Read BW Test
------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec]
65536 10000 1331.10 1329.85
------------------------------------------------------------------
------------------------------------------------------------------
RDMA_Write BW Test
------------------------------------------------------------------
#bytes #iterations BW peak[MB/sec] BW average[MB/sec]
65536 10000 1423.43 1422.64
------------------------------------------------------------------
2.) RDMA_CM + verbs: ib_rdma_bw -c -n100000
10711: Bandwidth peak (#0 to #7603): 1428.73 MB/sec
10711: Bandwidth average: 1428.13 MB/sec
3.) DAPL + RDMA_CM + verbs:
dapltest -T P -m p -s cst-50-ib0 -i 10000 RW 65535
RDMA_WRITES:
Total Time : 4.55 sec
Total Data Exchanged : 6249.90 MB
CPU Utilization : 25.30
Operation Throughput : 21952.66 ops/sec
Bandwidth : 1372.2 MB/sec
dapltest -T P -m p -s cst-50-ib0 -i 10000 RR 65535
RDMA_READS
Total Time : 4.67 sec
Total Data Exchanged : 6249.90 MB
CPU Utilization : 25.26
Operation Throughput : 21384.77 ops/sec
Bandwidth : 1336.52 MB/sec
More information about the general
mailing list