4 messages in com.perforce.perforce-user[p4] Network capacity recommendations...
FromSent OnAttachments
GSto...@cochlear.com.au10 Sep 2001 00:20 
Stephen Vance10 Sep 2001 00:45 
Arnt Gulbrandsen10 Sep 2001 05:16 
GSto...@cochlear.com.au11 Sep 2001 00:53 
Subject:[p4] Network capacity recommendations for perforce?
From:Arnt Gulbrandsen (ar@gulbrandsen.priv.no)
Date:09/10/2001 05:16:39 AM
List:com.perforce.perforce-user

GStoney at cochlear.com.au

Anyone out there willing to give me a recommendation on Perforce networking requirements between remote clients and server for capacity planning purposes?

Where I sit (Germany, despite my email address), there is one outstanding alternative:

Get 144kbps-both-ways DSL connections, proper ones with static IP addresses. Called "Business DSL" or similar. Set up a VPN. If the speed isn't good enough, it's really really easy to upgrade as far as 2.3Mbps.

I expect the high and low ends of the DSL spectrum are different in Australia, but there ought to be "business DSL" connections there too, where you can get some static addresses and a decent uplink as well as downlink.

I know this is an extremely open-ended question (obviously, faster is better), but what I'm after are some ballpark numbers for the point at which increasing bandwidth between 20 or so active developers and a remote server hits diminishing returns. I've read Technical Note 005 which purports to provide information on system requirements, but the section on Network requirements doesn't provide any meaningful information from a capacity planning point of view.

For example: - What is the maximum "ping" latency that you'd be happy to use P4Win with, before the GUI response time starts to become painful?

Not much more than 0.3 seconds. But I'm not at all tolerant of coffee-cup GUIs.

- What bandwidth per client would you plan for on an "average" project utilising extreme/rapid/incremental/concurrent/insert-best-practice-buzzword-of-the-week-here software development techniques?

Perforce is difficult that way... it uses very, very little bandwidth in the normal case, but the peaks sharp. It's really a question of how tolerant you are of peaks, and how often your developers act simultaneously, causing even worse usage peaks.

Initial sync is very heavy. Submits are heavy. Diffs are heavy. The rest is near zero.

--Arnt