v**n
发帖数: 951 | 1 DNS sucks应该没人反对吧?
papers倒是不少，能解决问题好像没多少。大家啥想法？展开说说。 |
s*****g
发帖数: 1055 | 2 You need to be more specific. DNS sucks in what way? what problem are you
trying to solve? People are using DNS every second, it can not be that bad. |
v**n
发帖数: 951 | 3 the way how the name is resolved. on one hand, if your local DNS can
cache everything, it is just a RTT. normally we can't. so there may be
multiple RTTs.
plus, if a record is changed, updating all the interested parties would
be a nightmare.
also, if nodes are on the move, the name/addr mapping could be changed
relatively often.
you
bad.
【在 s*****g 的大作中提到】
: You need to be more specific. DNS sucks in what way? what problem are you
: trying to solve? People are using DNS every second, it can not be that bad.
|
z**r
发帖数: 17771 | 4 what is a better way to maintain such a vast database?
【在 v**n 的大作中提到】
: the way how the name is resolved. on one hand, if your local DNS can
: cache everything, it is just a RTT. normally we can't. so there may be
: multiple RTTs.
: plus, if a record is changed, updating all the interested parties would
: be a nightmare.
: also, if nodes are on the move, the name/addr mapping could be changed
: relatively often.
:
: you
: bad.
|
s*****g
发帖数: 1055 | 5 The vast database needs to be distributed, no question about it. The problem
is the efficiency of new record propagation and existing record querying.
Current model to achieve this is pretty much ad hoc in application level.
If we can figure out a way to design a DNS system such that queries/propagations can be routed/summarized like IP has,that would drastically increase the scalability,
robustness, efficiency of the system. We can do L2 routing already, doing L7
routing is not totally out of q |
z**r
发帖数: 17771 | 6 跟俺昨天想到的几乎一样，后来自己给否了。原因是IP是二进制，长度也固定，非常容
易structural。而domain name长度不固定，如果都按照最大长度来，真正有效部分（
假设也有类似net mask的东西）还是没有本质区别。到最后，还是到了.com等root，跟
现在的似乎没有区别。
俺看现在的cache这种就挺好，需要改进的是dns server之间如何更有效率的刷新
database，说到底还是application level。当然这个东西绝对可以放到网络设备上，
其实很多公司也有相应的产品，比如cisco, f5等，但还是跑在linux上和普通的dns
server没有本质区别
problem
propagations can be routed/summarized like IP has,that would drastically
increase the scalability,
L7
generation DNS system, do not leave this task only to application developers.
【在 s*****g 的大作中提到】
: The vast database needs to be distributed, no question about it. The problem
: is the efficiency of new record propagation and existing record querying.
: Current model to achieve this is pretty much ad hoc in application level.
: If we can figure out a way to design a DNS system such that queries/propagations can be routed/summarized like IP has,that would drastically increase the scalability,
: robustness, efficiency of the system. We can do L2 routing already, doing L7
: routing is not totally out of q
|
v**n
发帖数: 951 | 7 老大，能不能请教一下csco和f5的那种产品可以对DNS有所提高？最近突然对DNS有兴趣
了。
drastically
【在 z**r 的大作中提到】
: 跟俺昨天想到的几乎一样，后来自己给否了。原因是IP是二进制，长度也固定，非常容
: 易structural。而domain name长度不固定，如果都按照最大长度来，真正有效部分（
: 假设也有类似net mask的东西）还是没有本质区别。到最后，还是到了.com等root，跟
: 现在的似乎没有区别。
: 俺看现在的cache这种就挺好，需要改进的是dns server之间如何更有效率的刷新
: database，说到底还是application level。当然这个东西绝对可以放到网络设备上，
: 其实很多公司也有相应的产品，比如cisco, f5等，但还是跑在linux上和普通的dns
: server没有本质区别
:
: problem
|
z**r
发帖数: 17771 | 8 算法提高估计不一定，主要是implementation方面。还是属于load balancer的概念？
看看cisco ACE或者f5 big-ip之类？
【在 v**n 的大作中提到】
: 老大，能不能请教一下csco和f5的那种产品可以对DNS有所提高？最近突然对DNS有兴趣
: 了。
:
: drastically
|
c*****i
发帖数: 631 | 9 load balancer用在dns server上一直都有把.以前在enterprise network里面还有个简
单的load balance方法,就是用/32的host route.
【在 z**r 的大作中提到】
: 算法提高估计不一定，主要是implementation方面。还是属于load balancer的概念？
: 看看cisco ACE或者f5 big-ip之类？
|
c*****i
发帖数: 631 | 10 这个同意,记得前几年有一次很大的network outage就是最上面一级的dns servers遭到
dos attack.
【在 z**r 的大作中提到】
: 跟俺昨天想到的几乎一样，后来自己给否了。原因是IP是二进制，长度也固定，非常容
: 易structural。而domain name长度不固定，如果都按照最大长度来，真正有效部分（
: 假设也有类似net mask的东西）还是没有本质区别。到最后，还是到了.com等root，跟
: 现在的似乎没有区别。
: 俺看现在的cache这种就挺好，需要改进的是dns server之间如何更有效率的刷新
: database，说到底还是application level。当然这个东西绝对可以放到网络设备上，
: 其实很多公司也有相应的产品，比如cisco, f5等，但还是跑在linux上和普通的dns
: server没有本质区别
:
: problem
|
z**r
发帖数: 17771 | 11 不是把dns server当成server farm然后用load balancer。是专门用到DNS上的比如
cisco ace global site selector。
【在 c*****i 的大作中提到】
: load balancer用在dns server上一直都有把.以前在enterprise network里面还有个简
: 单的load balance方法,就是用/32的host route.
|
