Hi folks. This message may be a bit verbose and not as techie as the ones I''ve seen in this list, but describing the scenario will save a lot of messages. Scenario: A building with 17 floors, each floor with 24 offices (totals 408 offices) connected to the backbone through a border switch (1 vlan for each office). The offices can lease bandwidth of 64k, 128k, 256k, 512k, 1M and 2Mbps, according to their needs. We have 3 links to the internet, 2Mbps each. Currently, we use cisco''s bbsm to handle the task of allocating the leased bandwidth for the vlans. Problem: bbsm never performed as expected. It freezes, disconnect users, reboots for itself leaving us in an awkward situation. After wasting precious time with cisco and it''s product, we decided to move to another solution and since the boss is an enthusiast of open source software, we decided to go for a linux based solution. We found that queueing discipline may be the solution. The question: are we correct, I mean is qdisc+htb the right thing to be used in such a scenario? Has anyone out there seen a linux box handling so many networks? As I go deeper in this subject, I will come to share my thoughts and doubts with you guys. Hope to hear from you. Have a good one. AL
> Scenario: > > A building with 17 floors, each floor with 24 offices (totals 408 > offices) connected to the backbone through a border switch (1 vlan for > each office). The offices can lease bandwidth of 64k, 128k, 256k, > 512k, 1M and 2Mbps, according to their needs. We have 3 links to the > internet, 2Mbps each. Currently, we use cisco''s bbsm to handle the > task of allocating the leased bandwidth for the vlans. > Problem: bbsm never performed as expected. It freezes, disconnect > users, reboots for itself leaving us in an awkward situation. After > wasting precious time with cisco and it''s product, we decided to move > to another solution and since the boss is an enthusiast of open source > software, we decided to go for a linux based solution. > We found that queueing discipline may be the solution. > The question: are we correct, I mean is qdisc+htb the right thing to > be used in such a scenario? Has anyone out there seen a linux box > handling so many networks? > As I go deeper in this subject, I will come to share my thoughts and > doubts with you guys. > > Hope to hear from you. Have a good one. > > > AL > _______________________________________________LARTC is being used in University environments with even higher bandwidth usage, so yes, you can do it. The multiple links may complicate things if you don''t want to explicitely assign a VLAN to a link. While balancing across the links is not that difficult, combining the balancing with shaping to guarantee x amount of bandwidth could be difficult. Do you want to simply use HTB to limit rates for each office? If so, this should work very well. You''ll just need to build the routes to specify which network goes to which internet connection. Or, you could use a separate box for each internet connection if you break out the VLANs earlier. FYI, I''m using a Pentium Pro 200 to shape traffic on a 3Mb/3Mb connection using CBQ for VoIP, Video, Citrix, and Bulk. HTH, -Ron
On Thu, 14 Jul 2005 10:14:37 -0300 Alvaro Motta <alvarolmmotta@gmail.com> wrote:>Hi folks.Hi,>This message may be a bit verbose and not as techie as the ones I''ve >seen in this list, but describing the scenario will save a lot of >messages.>Problem: bbsm never performed as expected. It freezes, disconnect >users, reboots for itself leaving us in an awkward situation.First of all although I personally also think cisco''s are not very good for other things than routing, experience tells me that problems are often caused by misconfiguration. I''d check out with an experienced cisco professional.>The question: are we correct, I mean is qdisc+htb the right thing to >be used in such a scenario? Has anyone out there seen a linux box >handling so many networks?I expect you want to use many htb classes with the same parent. I have seen a reports that this causes problems when you really have a large class number (several hundred), but again, this may have been caused by misconfiguration. If you adapt your requirements however (every IP is handled equally), you can use WRR, it has been proven to work without problems under even larger number of clients (1400) and bandwidth (16Mbit). For an example see here: http://mailman.ds9a.nl/pipermail/lartc/2005q2/016500.html If you don''t want to adapt, once upon a time I wrote a management tool for an ISP with requirements similar to yours. Although I tuned it for performance and it seems to work well, as far as I know there are only a couple of dozen users, I don''t know how it would behave if it was used with several hunderd users.>ALYours sincerely, Peter
On Thu, 14 Jul 2005 17:24:00 +0200 Peter Surda <surda@shurdix.com> wrote:>If you don''t want to adapt, once upon a time I wrote a management tool for an >ISP with requirements similar to yours. Although I tuned it for performance and >it seems to work well, as far as I know there are only a couple of dozen users, >I don''t know how it would behave if it was used with several hunderd users.Update: I obtained some data from the mentioned ISP: - Backbone: 16Mbit - average transfer 700kB/s (5600kbit/s) - about 20 users - CPU Celeron 333 - no performance problems noticeable The only unanswered question remains the user count (20 vs 400 is not really comparable). Yours sincerely, Peter
Hello I have 40Mbit/s internet uplink Average transfer 25Mbit/s - 3957 users On machine: - CPU 3.2GHz P4 +HT enabled - 2GB RAM - Intel GB cards Machine is doing nat/dnat for each user so there is 3957 DNAT/SNAT ip pairs (private to public and vice versa) Machine load is 0.1 avg And no any problems :) On Thu, 14 Jul 2005 17:24:00 +0200 Peter Surda <surda@shurdix.com> wrote:>If you don''t want to adapt, once upon a time I wrote a management tool for an >ISP with requirements similar to yours. Although I tuned it for performance and >it seems to work well, as far as I know there are only a couple of dozen users, >I don''t know how it would behave if it was used with several hunderd users.Update: I obtained some data from the mentioned ISP: - Backbone: 16Mbit - average transfer 700kB/s (5600kbit/s) - about 20 users - CPU Celeron 333 - no performance problems noticeable The only unanswered question remains the user count (20 vs 400 is not really comparable). Yours sincerely, Peter
On Fri, 15 Jul 2005 11:55:34 +0200 "Pawe³ Staszewski" <pstaszewski@artcom.pl> wrote:>HelloHello>I have 40Mbit/s internet uplink >Average transfer 25Mbit/s >- 3957 users[cut] Well, the question is what kind of TC-setup you have. Do you have a separate HTB class for every user? Yours sincerely, Peter