z**r
发帖数: 17771 | 1 一直没太关注这家公司，但是最近发现客户用了很多这家公司的东西。 |
b******s
发帖数: 5329 | 2 a10攻击机？这东西基本打不烂。不卖中国的。。。。
玩笑话。
觉得他家的产品还行，好像是中国人开的？ |
b***p
发帖数: 700 | 3 你客户什么时候买的啊？好用吗？股票是烂的不能再烂了，他们过去2-3年加入的员工
基本上全部套牢
听说是特便宜
【在 z**r 的大作中提到】
: 一直没太关注这家公司，但是最近发现客户用了很多这家公司的东西。
|
b***p
发帖数: 700 | 4 主要是Foundry一票人出来干的，因为这个还要赔brocade -100M美刀，说人家偷专利/
技术
【在 b******s 的大作中提到】
: a10攻击机？这东西基本打不烂。不卖中国的。。。。
: 玩笑话。
: 觉得他家的产品还行，好像是中国人开的？
|
s*****g
发帖数: 1055 | 5 I wouldn't bet my retirement money on loadbalancer vendors, those boxes are
PITA to manage and troubleshoot, expensive, do not scale well, most web 2.0
companies do not use dedicated commercial load balancers, in stead, load
balancing can be achieved in other easier to manage and more scalable ways,
for example, building load balancing to applications layer coupled with
distributed systems. |
z**r
发帖数: 17771 | 6 主要去年的时候吧，换掉好多F5的东西。他家股票怎么回事儿，好像去年10月份大跌，
会有起色吗？抄底买些？
【在 b***p 的大作中提到】
: 你客户什么时候买的啊？好用吗？股票是烂的不能再烂了，他们过去2-3年加入的员工
: 基本上全部套牢
: 听说是特便宜
|
z**r
发帖数: 17771 | 7 application layer load balancing is not flexible? everytime you add some
capacity, you'll need to modify the code or application setup? cannot
imagine this is a better solution tho
are
0
,
【在 s*****g 的大作中提到】
: I wouldn't bet my retirement money on loadbalancer vendors, those boxes are
: PITA to manage and troubleshoot, expensive, do not scale well, most web 2.0
: companies do not use dedicated commercial load balancers, in stead, load
: balancing can be achieved in other easier to manage and more scalable ways,
: for example, building load balancing to applications layer coupled with
: distributed systems.
|
s*****g
发帖数: 1055 | 8 My point was that you can build applications like the way how GFS/hadoop
operates, you have a "head node" to handle all requests but the distribution
algorithm of load is not on IP layer -- in a sense that this head node has
more knowledge of the worker nodes than a traditional load balancer does.
With this approach, you take the complexity out of underlying networking
transport, hence better control of your destiny.
I have to admit that I don't have extensive knowledge operating load
balancers, but enough to always avoid using them in my line of work.
【在 z**r 的大作中提到】
: application layer load balancing is not flexible? everytime you add some
: capacity, you'll need to modify the code or application setup? cannot
: imagine this is a better solution tho
:
: are
: 0
: ,
|
z**r
发帖数: 17771 | 9 this sounds like a re-invention of the wheels, a story of 20 years ago. :)
SLB can do DPI to understand the applications but with better scale and more
standard setup. of coz you can always customize the solution as well.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
s*****g
发帖数: 1055 | 10 It is not reinventing, and it has nothing to do with DPI, the head node has
more knowledge of the workers in a sense that they speak their own
application level protocols, i.e. Zookeeper etc
more
【在 z**r 的大作中提到】
: this sounds like a re-invention of the wheels, a story of 20 years ago. :)
: SLB can do DPI to understand the applications but with better scale and more
: standard setup. of coz you can always customize the solution as well.
:
: distribution
: has
|
|
|
b***p
发帖数: 700 | 11 one key benefit of hardware load balancer is the ssl offload, which can use
silicon instead of expensive cpu cycles. another one is ddos protection, you
don't want to expose your sever gears to world.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
s*****g
发帖数: 1055 | 12 CPU cycles are really cheap, off cause you will have some network devices to
filter out illegal traffic, how can load balancer defeat ddos attack in any
ways？ |
b***p
发帖数: 700 | 13 general cpu is not good at descryption kind of jobs. actually a10 ddos
device is based on its lb product, maybe on diff boxes. but it just filter
out those suspicious req based on fingerprint.
to
any
【在 s*****g 的大作中提到】
: CPU cycles are really cheap, off cause you will have some network devices to
: filter out illegal traffic, how can load balancer defeat ddos attack in any
: ways？
|
L******t
发帖数: 1985 | 14 What you said is only applicable to a chosen few hi-tech companies, like
Google, Amazon etc. Most companies are not capable of that.
Load balancer is kinda like security device, focusing on what happen within
a machine. While network engineers focus on what happen between machines.
Different kinds of people/job.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
w*f
发帖数: 111 | 15 It's good alternative to F5. They're cheaper and price includes all features
. For enterprise, there is a lot of need for commercial SLB instead of home-
grown solution. Who is gonna support it? |
z**r
发帖数: 17771 | 16 slb is a dpi product
to
any
【在 s*****g 的大作中提到】
: CPU cycles are really cheap, off cause you will have some network devices to
: filter out illegal traffic, how can load balancer defeat ddos attack in any
: ways？
|
s*****g
发帖数: 1055 | 17 小公司更不需要SLB了,一些开源软件(HAproxy etc)足够满足需求了。现在网络工程师
的职责不能只局限于机器机器之间的通信了，网络是整个技术operation 不可分割的一
部分，网络需求直接从应用层工程师,系统管理工程师来，当然还要考虑 cost。
within
【在 L******t 的大作中提到】
: What you said is only applicable to a chosen few hi-tech companies, like
: Google, Amazon etc. Most companies are not capable of that.
: Load balancer is kinda like security device, focusing on what happen within
: a machine. While network engineers focus on what happen between machines.
: Different kinds of people/job.
:
: distribution
: has
|
w*f
发帖数: 111 | 18 小公司has little IT budget so they can use 开源软件.
SP like Google has the resource to develop in house or use 开源软件;
For most companies, who is gonna support HAproxy 2 am in the morning?
tech support is very important for enterprise customers who like Cisco for a
reason. |
z**r
发帖数: 17771 | 19 一直没太关注这家公司，但是最近发现客户用了很多这家公司的东西。 |
b******s
发帖数: 5329 | 20 a10攻击机？这东西基本打不烂。不卖中国的。。。。
玩笑话。
觉得他家的产品还行，好像是中国人开的？ |
|
|
b***p
发帖数: 700 | 21 你客户什么时候买的啊？好用吗？股票是烂的不能再烂了，他们过去2-3年加入的员工
基本上全部套牢
听说是特便宜
【在 z**r 的大作中提到】
: 一直没太关注这家公司，但是最近发现客户用了很多这家公司的东西。
|
b***p
发帖数: 700 | 22 主要是Foundry一票人出来干的，因为这个还要赔brocade -100M美刀，说人家偷专利/
技术
【在 b******s 的大作中提到】
: a10攻击机？这东西基本打不烂。不卖中国的。。。。
: 玩笑话。
: 觉得他家的产品还行，好像是中国人开的？
|
s*****g
发帖数: 1055 | 23 I wouldn't bet my retirement money on loadbalancer vendors, those boxes are
PITA to manage and troubleshoot, expensive, do not scale well, most web 2.0
companies do not use dedicated commercial load balancers, in stead, load
balancing can be achieved in other easier to manage and more scalable ways,
for example, building load balancing to applications layer coupled with
distributed systems. |
z**r
发帖数: 17771 | 24 主要去年的时候吧，换掉好多F5的东西。他家股票怎么回事儿，好像去年10月份大跌，
会有起色吗？抄底买些？
【在 b***p 的大作中提到】
: 你客户什么时候买的啊？好用吗？股票是烂的不能再烂了，他们过去2-3年加入的员工
: 基本上全部套牢
: 听说是特便宜
|
z**r
发帖数: 17771 | 25 application layer load balancing is not flexible? everytime you add some
capacity, you'll need to modify the code or application setup? cannot
imagine this is a better solution tho
are
0
,
【在 s*****g 的大作中提到】
: I wouldn't bet my retirement money on loadbalancer vendors, those boxes are
: PITA to manage and troubleshoot, expensive, do not scale well, most web 2.0
: companies do not use dedicated commercial load balancers, in stead, load
: balancing can be achieved in other easier to manage and more scalable ways,
: for example, building load balancing to applications layer coupled with
: distributed systems.
|
s*****g
发帖数: 1055 | 26 My point was that you can build applications like the way how GFS/hadoop
operates, you have a "head node" to handle all requests but the distribution
algorithm of load is not on IP layer -- in a sense that this head node has
more knowledge of the worker nodes than a traditional load balancer does.
With this approach, you take the complexity out of underlying networking
transport, hence better control of your destiny.
I have to admit that I don't have extensive knowledge operating load
balancers, but enough to always avoid using them in my line of work.
【在 z**r 的大作中提到】
: application layer load balancing is not flexible? everytime you add some
: capacity, you'll need to modify the code or application setup? cannot
: imagine this is a better solution tho
:
: are
: 0
: ,
|
z**r
发帖数: 17771 | 27 this sounds like a re-invention of the wheels, a story of 20 years ago. :)
SLB can do DPI to understand the applications but with better scale and more
standard setup. of coz you can always customize the solution as well.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
s*****g
发帖数: 1055 | 28 It is not reinventing, and it has nothing to do with DPI, the head node has
more knowledge of the workers in a sense that they speak their own
application level protocols, i.e. Zookeeper etc
more
【在 z**r 的大作中提到】
: this sounds like a re-invention of the wheels, a story of 20 years ago. :)
: SLB can do DPI to understand the applications but with better scale and more
: standard setup. of coz you can always customize the solution as well.
:
: distribution
: has
|
b***p
发帖数: 700 | 29 one key benefit of hardware load balancer is the ssl offload, which can use
silicon instead of expensive cpu cycles. another one is ddos protection, you
don't want to expose your sever gears to world.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
s*****g
发帖数: 1055 | 30 CPU cycles are really cheap, off cause you will have some network devices to
filter out illegal traffic, how can load balancer defeat ddos attack in any
ways？ |
|
|
b***p
发帖数: 700 | 31 general cpu is not good at descryption kind of jobs. actually a10 ddos
device is based on its lb product, maybe on diff boxes. but it just filter
out those suspicious req based on fingerprint.
to
any
【在 s*****g 的大作中提到】
: CPU cycles are really cheap, off cause you will have some network devices to
: filter out illegal traffic, how can load balancer defeat ddos attack in any
: ways？
|
L******t
发帖数: 1985 | 32 What you said is only applicable to a chosen few hi-tech companies, like
Google, Amazon etc. Most companies are not capable of that.
Load balancer is kinda like security device, focusing on what happen within
a machine. While network engineers focus on what happen between machines.
Different kinds of people/job.
distribution
has
【在 s*****g 的大作中提到】
: My point was that you can build applications like the way how GFS/hadoop
: operates, you have a "head node" to handle all requests but the distribution
: algorithm of load is not on IP layer -- in a sense that this head node has
: more knowledge of the worker nodes than a traditional load balancer does.
: With this approach, you take the complexity out of underlying networking
: transport, hence better control of your destiny.
: I have to admit that I don't have extensive knowledge operating load
: balancers, but enough to always avoid using them in my line of work.
|
w*f
发帖数: 111 | 33 It's good alternative to F5. They're cheaper and price includes all features
. For enterprise, there is a lot of need for commercial SLB instead of home-
grown solution. Who is gonna support it? |
z**r
发帖数: 17771 | 34 slb is a dpi product
to
any
【在 s*****g 的大作中提到】
: CPU cycles are really cheap, off cause you will have some network devices to
: filter out illegal traffic, how can load balancer defeat ddos attack in any
: ways？
|
s*****g
发帖数: 1055 | 35 小公司更不需要SLB了,一些开源软件(HAproxy etc)足够满足需求了。现在网络工程师
的职责不能只局限于机器机器之间的通信了，网络是整个技术operation 不可分割的一
部分，网络需求直接从应用层工程师,系统管理工程师来，当然还要考虑 cost。
within
【在 L******t 的大作中提到】
: What you said is only applicable to a chosen few hi-tech companies, like
: Google, Amazon etc. Most companies are not capable of that.
: Load balancer is kinda like security device, focusing on what happen within
: a machine. While network engineers focus on what happen between machines.
: Different kinds of people/job.
:
: distribution
: has
|
w*f
发帖数: 111 | 36 小公司has little IT budget so they can use 开源软件.
SP like Google has the resource to develop in house or use 开源软件;
For most companies, who is gonna support HAproxy 2 am in the morning?
tech support is very important for enterprise customers who like Cisco for a
reason. |
