On Tue, 28 Nov 2000, mohan kanumuru wrote:
> Is it a good step to increase number of virtual
> interfaces/robots to 1000/machine and keep the request
> rate/robot to 0.4/sec. Will this help to achive
> 400requests/sec/machine without experiencing "virtual
> memory exceeded in new" problem and help not generate
> a best-effort workload.
There are several semi-orthogonal issues here.
Under normal conditions, keeping maximum rate/robot at 0.4/sec will
ensure that the workload is not a best-effort one.
Increasing the number of robots to 1000/machine and keeping maximum
rate/robot at 0.4/sec is one possible way of achieving 400/sec load
provided your environment (OS, hardware, network gear, etc.) can
support the load. Doing so keeps the workload similar to PolyMix.
There are other ways of achieving the same result (e.g., removing or
increasing the open_conn_lmt).
Whether you use 1000 IP addresses or one IP address for all robots is
a separate issue affecting the size of ARP tables only.
******
When you are writing a custom workload file, try to think of the
real-life objects you are describing. There is little magic behind PGL
objects. Most of them are a direct reflection of real world entities.
PGL is a double-edged sword: PGL approaches (and in some cases
exceeds) the flexibility of real configurations within the problem
domain, and/but/while giving you plenty of room for mistakes, just
like in real life...
HTH,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:16 MDT