New network bandwidth requirements | Unix Linux Forums | IP Networking

  Go Back    


IP Networking Learn TCP/IP, Internet Protocol, Routing, Routers, Network protocols in this UNIX and Linux forum.

New network bandwidth requirements

IP Networking


Tags
bandwidth, ip, provisioning, tcp

Closed Thread    
 
Thread Tools Search this Thread Display Modes
    #1  
Old 06-07-2013
redrider redrider is offline
Registered User
 
Join Date: Jun 2013
Last Activity: 13 June 2013, 9:57 AM EDT
Posts: 1
Thanks: 0
Thanked 0 Times in 0 Posts
New network bandwidth requirements

Many papers, articles and posts about network bandwidth requirements refer to network traffic rules-of-thumb when estimating bandwidth requirements for a new network with an unknown load. I've seen a couple rules for video and VoIP.

I'd be thankful if someone could share any rules-of thumb they know of.

Thanks!!!
Sponsored Links
    #2  
Old 06-11-2013
DGPickett DGPickett is offline Forum Advisor  
Registered User
 
Join Date: Oct 2010
Last Activity: 21 November 2014, 7:36 PM EST
Location: Southern NJ, USA (Nord)
Posts: 4,522
Thanks: 8
Thanked 554 Times in 531 Posts
More than enough beats less? Never overlook compressed tunnels in a pinch. Unknown in means unknown out?
Sponsored Links
    #3  
Old 06-11-2013
Corona688 Corona688 is offline Forum Staff  
Mead Rotor
 
Join Date: Aug 2005
Last Activity: 21 November 2014, 5:42 PM EST
Location: Saskatchewan
Posts: 19,902
Thanks: 845
Thanked 3,413 Times in 3,198 Posts
It's certainly easier to downgrade than upgrade.
    #4  
Old 06-12-2013
DGPickett DGPickett is offline Forum Advisor  
Registered User
 
Join Date: Oct 2010
Last Activity: 21 November 2014, 7:36 PM EST
Location: Southern NJ, USA (Nord)
Posts: 4,522
Thanks: 8
Thanked 554 Times in 531 Posts
If you have too much you can measure how much less you could do with, but if you have too little, how much too little gets buried in backed up and even dropped packets. However, the usual rules about service length and service arrival frequency versus number of servers, at a very high level, applies. Everything is a server at a high level: a CPU, an ethernet link, a disk controller, a disk drive, a fiber link. With one server, it is very easy for backlog to occur even if the averages are not violated. Think snake that swallowed a basketball. As you approach critical loading, you will see backlog peaks start occurring due to periodic bad luck. With one server, like one line at the bank, if you can remember when people walked to the bank, one customer with big needs makes a huge line that lasts for hours.
Sponsored Links
    #5  
Old 06-20-2013
Just Ice's Avatar
Just Ice Just Ice is offline Forum Advisor  
Lights on, brain off.
 
Join Date: Mar 2005
Last Activity: 20 September 2013, 10:59 AM EDT
Location: Philadelphia metro
Posts: 958
Thanks: 3
Thanked 67 Times in 60 Posts
ever heard of the new highway? some people built a new highway because the old road was getting too small for the amount of traffic on it. so everybody was happy when the new highway was done. while the highway was still new, everybody said it was great that they had that much space to run and cruise in it. so everybody's friends and families soon started driving on the new highway until eventually everybody was thinking the highway was now too small and they needed another new highway ...

moral of the story? if you do not have enough, you will soon be looking for some more. if you have more than enough, people will start filling it up sooner or later.


as i almost always had to guesstimate things because of the lack of access to network metrics, i look at current load and whether that load has increased over the years as well as whether there are plans to put in more network-hungry apps/devices (i.e., video conferencing). if current load has been increasing, i look at the amount and rate of the increase.

as an example: 10 users at a remote site using 5 applications at the corporate site at year 1 and 50 users (max capacity) at the remote site using same 5 applications at the corporate site at year 5 -- only the user base increased. minus any planned implementation of network-hungry apps/devices and no complaints of performance hits even with the increase in the user load, i will not be looking to upgrade within the next 2 years unless the company just wanted to spend money just because.

for a site with an unknown load, look at the kind and size of the population the site most likely will be hosting as well as the apps/devices they will be using and the planned infrastructure physically located at the site. if you have no other site that is currently comparable to the new site, check how many seats are planned in the space as well as how big is the data center/room and correlate that with the business and/or group that will be located there (i.e., software development, banking back-office, internet research, etc.). decide the approximate load and plan accordingly. if you are still at a loss and still cannot decide, ask the project manager how much they have budgeted for the whole project and just reasonably ask for as much as they can give you.
Sponsored Links
    #6  
Old 06-20-2013
DGPickett DGPickett is offline Forum Advisor  
Registered User
 
Join Date: Oct 2010
Last Activity: 21 November 2014, 7:36 PM EST
Location: Southern NJ, USA (Nord)
Posts: 4,522
Thanks: 8
Thanked 554 Times in 531 Posts
Parkinson's law: need expands to consume excess resources.

Companies like Akamai make a good living ensuring your static web hits are filled from relatively local cache servers.

Good architectural design has to deal with:
  • having a soft saturation, so throughput goes up to saturation and then excess load is shedded in a least-lost-value basis, like newest clients lower in priority than older clients (deeper into transaction process).
  • avoiding negative saturation behaviors like overloaded Ethernet, which actually slows down due to collisions creating lost time on wire. Positive saturation behavior means the higher the overload, the more efficient the process. Requests can be sorted to they have higher locality of reference. Sometimes, requests for the same file can be mbone multicast as one. Sorting by disk position means shorter seeks.
  • flow control mechanisms allow services beyond capacity to be queued for eventual fulfillment, but service cancellation is quickly forearded to the server. The bad behaviors are thing like sending service requests every n seconds until a reply is received, consuming precious bandwidth and cluttering the server with cancelled, redundant, prior requests. Some routers can stifle keep-alive traffic, say from tcp connections of queued services, so they do not drag down net speed.
Sponsored Links
Closed Thread

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

More UNIX and Linux Forum Topics You Might Find Helpful
Thread Thread Starter Forum Replies Last Post
Bandwidth montor palm101 IP Networking 4 05-06-2008 03:48 PM
bandwidth big123456 IP Networking 4 12-07-2006 02:29 AM
Bandwidth monitoring Crazy_murli UNIX for Advanced & Expert Users 4 07-21-2006 06:53 AM
Bandwidth Caping With PF nocturnal IP Networking 2 10-15-2003 04:54 PM
bandwidth check? solvman UNIX for Dummies Questions & Answers 3 01-31-2002 05:21 PM



All times are GMT -4. The time now is 04:36 AM.