Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hello, On Wed, 9 Apr 2008 10:49:57 +0200 Ralf Gross wrote: > Christian Balzer schrieb: > > > If your local I/O subsystem pulls 120MB/s, the expected max DRBD > > > throughput is around 105 MB/s: > > > > > > - disk does 120, > > > - Gigabit Ethernet realistically does 110, > > Yeah, various network tests get near that value. > > > > > - so the network is your bottleneck, > > It's a bonded dual GigE. Initially done to give DRBD enough breathing > > room and left in place to have more redundancy for it and heartbeat. > > Are the servers directly connected or are you using switch to connect? Directly. > I tried the round-robin bonding mode with cisco switches, but cisco > doesn't support rr mode. At least when I tested it. So in the end the > connection between my 2 server was still limited by a single GbE link. > Yeah, I heard that from somebody else before, too. But no need (distance/topology wise) to involve switches and thus more possible points of failure in my case, so direct connections it is. > In a test with 2 x GbE crossover connections, I was able to achive > ~170 MB/s with the netpipe benchmark. I didn't test this with drbd, > but other daemons (smbd, vsftp) didn't show a huge impovement from his > raw speed gain. The limit was always ~80 MB/s. > That netpipe result matches mine here, I think I used a proftpd server for raw transfer tests (with ssh/scp the CPU was always the limiting factor) and it was definitely over the speed of a single link. But then again these machines have 24GB RAM, so could transfer the 8GB test file without ever hitting the disks. So maybe your 80MB/s are more related to your storage system bandwidth? Gruesse aus Tokio, Christian -- Christian Balzer Network/Systems Engineer NOC chibi at gol.com Global OnLine Japan/Fusion Network Services http://www.gol.com/