1 Sep 2009 23:07
testing network latency
Seth Galitzer <sgsax <at> ksu.edu>
2009-09-01 21:07:47 GMT
2009-09-01 21:07:47 GMT
I've got 24 Sunray 2FS DTUs hanging off a single SRSS server (whitebox, 2x Xeon 2.6GHz CPU, 6GB RAM, Solaris 10). By default, they connect to a Windows Terminal Server (SunFire X4600, 4x Opteron 8218 2.6GHz CPU, 16GB RAM, Windows Server 2003) via srwc. The DTUs are on a completely separate subnet from the rest of my network. The SRSS host has a single interface that is on this subnet that is used for all DTU communication. The SRSS host also has a direct cable link (crossover cable) to the WTS host. Both the SRSS host interfaces and the WTS interface are all 1Gbps. This setup is used in a classroom lab setting. During class, there can be anywhere from 10-20 students using the lab at the same time. We are a computer science program, so students use development platform applications, such as Eclipse, Visual Studio, and Rational System Architect (formerly Rational Rose), among others. In my second year of running this system, we seem to be having significant performance problems with a large number of users trying to load applications simultaneously. eg, if a class comes in and they all login and try to load eclipse, the system will grind to a halt and become completely unresponsive until the app is loaded for everybody. This can take up to 30 min before the DTUs are usable. I've been monitoring resource usage on the WTS host using Sysinternals Process Explorer (http://technet.microsoft.com/en-us/sysinternals/bb896653.aspx). As far as I can tell when the system gets bogged down like this, I am not seeing any CPU. memory, or I/O wait causing the latency. The CPU utilization jumps around between 20-60%, and I only see usage spikes on 2 of the 8 CPU cores. RAM utilization stays absolutely stable, right around 4GB of my 16 available. File I/O is barely noticeable.(Continue reading)
RSS Feed