Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Nobody who had to work with 6509's/7609's at an ISP is shedding a tear over this.

Someone (Richard Steenbergen?) once made a joke that we should take the last 6509/7609 and launch it into orbit to celebrate.

It's not that they weren't popular. At one point in the mid 2000's they appeared to make up about 1/3 of major internet routers (if you looked around a carrier hotel). This was due to their extremely low cost compared to actual high end routers. While they had serious limitations and were notoriously sensitive to "IOS roulette", somehow you could just make them work.



The 7600 was an absolutely idiotic product. The 6500 was, for the time, fine as an enterprise Ethernet switch (much more capable obviously once the sup 2 with fabric services module and sup 720 with integrated crossbar came along,) but using it as a ISP router, especially where you were taking a full routing table? That was just stupid.

For anyone reading this that doesn't have experience with these things, when the parent commenter talks about "just making them work," one failure mode among many in these devices is that packet forwarding is primarily done in hardware, more or less at line rate. But, if you enable an IOS feature that isn't supported in hardware, it gets processed in software. In more "ISP-focused" routers, it is common to just not support features that aren't implemented in hardware. Forwarding performance on these platforms goes from almost 500 million packets per second in hardware (in certain highly specific and very unlikely scenarios) to around 40 - 50 thousand packets per second -- absolute best-case -- in software. Another failure mode specifically applicable to the ISP scenario is the fixed hardware forwarding table size, which for many models was 192k IPv4 prefixes. could you have a larger forwarding table size? Absolutely. In software.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: