| 1 |
Did you ask your question/search on the Beowulf ML? This one seems familiar and is a |
| 2 |
recurrent theme on that ML. |
| 3 |
|
| 4 |
Eric |
| 5 |
|
| 6 |
> I've got the following system: |
| 7 |
> |
| 8 |
> Linux server 2.6.26 #5 SMP Fri Feb 6 12:18:54 CST 2009 ppc64 PPC970FX, |
| 9 |
> altivec supported RackMac3,1 GNU/Linux |
| 10 |
> |
| 11 |
> Setup on a cisco gig switch with both gigabit NICs in the following setup: |
| 12 |
> |
| 13 |
> % cat /proc/net/bonding/bond0 |
| 14 |
> Ethernet Channel Bonding Driver: v3.2.5 (March 21, 2008) |
| 15 |
> |
| 16 |
> Bonding Mode: adaptive load balancing |
| 17 |
> Primary Slave: None |
| 18 |
> Currently Active Slave: eth0 |
| 19 |
> MII Status: up |
| 20 |
> MII Polling Interval (ms): 100 |
| 21 |
> Up Delay (ms): 0 |
| 22 |
> Down Delay (ms): 0 |
| 23 |
> |
| 24 |
> Slave Interface: eth0 |
| 25 |
> MII Status: up |
| 26 |
> Link Failure Count: 0 |
| 27 |
> Permanent HW addr: 00:0d:93:9e:2b:ca |
| 28 |
> |
| 29 |
> Slave Interface: eth1 |
| 30 |
> MII Status: up |
| 31 |
> Link Failure Count: 0 |
| 32 |
> Permanent HW addr: 00:0d:93:9e:2b:cb |
| 33 |
> |
| 34 |
> Both eth0 and eth1 are running at 1000Mb/s full according to ethtool. |
| 35 |
> I start netserver (from netperf) on the server and then run netperf on |
| 36 |
> four other singlegigabit connected clients on the same switch with: |
| 37 |
> |
| 38 |
> netperf -l 45 -t TCP_STREAM -H server |
| 39 |
> |
| 40 |
> but I get the following outputs: |
| 41 |
> |
| 42 |
> client 1: |
| 43 |
> |
| 44 |
> Recv Send Send |
| 45 |
> Socket Socket Message Elapsed |
| 46 |
> Size Size Size Time Throughput |
| 47 |
> bytes bytes bytes secs. 10^6bits/sec |
| 48 |
> |
| 49 |
> 87380 16384 16384 45.04 405.68 |
| 50 |
> |
| 51 |
> client 2: |
| 52 |
> |
| 53 |
> Recv Send Send |
| 54 |
> Socket Socket Message Elapsed |
| 55 |
> Size Size Size Time Throughput |
| 56 |
> bytes bytes bytes secs. 10^6bits/sec |
| 57 |
> |
| 58 |
> 87380 16384 16384 45.01 216.07 |
| 59 |
> |
| 60 |
> client 3: |
| 61 |
> |
| 62 |
> Recv Send Send |
| 63 |
> Socket Socket Message Elapsed |
| 64 |
> Size Size Size Time Throughput |
| 65 |
> bytes bytes bytes secs. 10^6bits/sec |
| 66 |
> |
| 67 |
> 87380 16384 16384 45.87 143.93 |
| 68 |
> |
| 69 |
> client 4: |
| 70 |
> |
| 71 |
> Recv Send Send |
| 72 |
> Socket Socket Message Elapsed |
| 73 |
> Size Size Size Time Throughput |
| 74 |
> bytes bytes bytes secs. 10^6bits/sec |
| 75 |
> |
| 76 |
> 87380 16384 16384 45.05 164.05 |
| 77 |
> |
| 78 |
> adding those up gives me: |
| 79 |
> |
| 80 |
> $ echo "405.68+216.07+143.93+164.05" | bc -l |
| 81 |
> 929.73 |
| 82 |
> |
| 83 |
> Not more than 1gbps. Shouldn't I be able to get >> 1gbps with balance-alb mode? |
| 84 |
> |
| 85 |
> |
Archives