Does anyone have experience with using HTB to shape traffic for several thousands of users, each with a class of their own? The LARTC HOWTO has a section about setting up hashed filters to deal with "many" classes. Has anyone used that setup? -- ams
Hi,>Does anyone have experience with using HTB to shape traffic for several >thousands of users, each with a class of their own? The LARTC HOWTO has >a section about setting up hashed filters to deal with "many" classes. >Has anyone used that setup?I''m using HTB to shape traffic generated by ~3000 users using hashing filters. It works OK and is cpu conservative (thanks to hashing). One thing is worth mentioting if you''re going to update shaping config more often. Load tc commands using "batch" option (-b) of tc binary instead of sequential executions of tc binary. I''ll load your config in ~30 seconds instead of couple of minutes. I''m curious about one thing. What kind of hardware are you planning to use for that router? cheers, Marek Kierdelewicz KoBa ISP
Hi Marek. At 2007-10-03 14:44:29 +0200, marek@piasta.pl wrote:> > I''m using HTB to shape traffic generated by ~3000 users using hashing > filters. It works OK and is cpu conservative (thanks to hashing).Thanks, that''s useful to know.> Load tc commands using "batch" option (-b) of tc binaryYes, that''s a very good point. (I already do this.)> I''m curious about one thing. What kind of hardware are you planning to > use for that router?Right now we have one HP ProLiant DL380 G5 machine, with a 2GHz Xeon processor and 2GB of memory. Part of the reason I posted was because I wanted to get a better idea of how much CPU/memory I would need for a setup that needed to handle some 15k classes with bandwidth limits of ~64kbit each. What hardware are you using for your 3k-user router? -- ams