Performance KPI for an ISP

[Zarrar Hasham Khan]

Dear TGP Team,

My company (not in Pakistan) which is a mobile operator is currently in
talks with a new ISP to lease about 2.5Gbps bandwidth. Right now the
price we are getting is quite competitive compared to other ISP we have
(We have currently about 1.5Gbps each with two ISP). Even though there
is a great benefit for cost optimization if we go with this ISP but I am
afraid there may be compromise in the quality.

I am coming from Mobile Network background and I do not have as yet
in-depth understanding of the ISP Performance side of things. We are
going to sign a contract with them and a certain component will relate
to the KPI & SLA. I need your help in understanding what are the key KPI
(and thresholds) that can measure the quality of bandwidth we are
getting from this ISP. We will put them in the SLA.

We currently do not have any serious KPI based SLA with the other two
operators as the services we are getting from them are very good (but we
get them at a price premium). Any help will be greatly appreciated.

Regards,

Zarrar

[Haris Shamsi]

Zarrar You can refer to the same GSMA IR 3444 document for data
Connectivity for delay, jitters etc.

While doing peering with two and loadbalancing the traffic, TCP
fragmentation is one of the consideration, which I am sure will be
managed by your nodes.

Your BW provider needs to have good (efficient path to destination),
so while doing an SLA with them don't measure only the last mile but
their peering points and strengths also.

I should be able to send you some standard SLAs offline but need to
dig it from my backup archive. Will take some time ...

/HS

> --
> Manage your subscription:
> http://groups.google.com/group/telecom-grid-pakistan
> Browse and populate the community wiki at: http://wiki.telecompk.net

--
Sent from my mobile device

*http://cloud.pk*
*
*You're either part of the solution or part of the problem - Eldridge
Cleaver

[Majid Farid]

Just on the same point since you are a service provider it might not
be good idea to go with only with on provider. Cost might be the key
decision factor however if you have diverse link on different provider
it will help ensure better "customer experience" :)

Having said that, Ericsson manages Sprint network in US which amounts
for 20% of Internet traffic. All trans-Atlantic SLA can be viewed on
https://www.sprint.net/sla_performance.php .. Sprintlink is our
Internet transit pipe.

On Apr 8, 11:13 am, Haris Shamsi <haris.sha...@gmail.com> wrote:
> Zarrar You can refer to the same GSMA IR 3444 document for data
> Connectivity for delay, jitters etc.
>
> While doing peering with two and loadbalancing the traffic, TCP
> fragmentation is one of the consideration, which I am sure will be
> managed by your nodes.
>
> Your BW provider needs to have good (efficient path to destination),
> so while doing an SLA with them don't measure only the last mile but
> their peering points and strengths also.
>
> I should be able to send you some standard SLAs offline but need to
> dig it from my backup archive. Will take some time ...
>
> /HS
>

[Zarrar Khan]

Hi TGP,

Has anyone tried Google Global Cache? How is the youtube cache performance? What sort of startup delay have you experienced?

We have been contacted by Google to host the cache. They will source all hardware and servers. We need to provide space, power and subnet. Primary caching will be for youtube videos.

Is it worth the hassle? We already have a cache on our Gi.

Cheers,

Zarrar

[Tee Emm]

I am not sure about improvement in the startup delay but the amount of transit bandwidth it could save on popular contents could be one big plus to having this set up in place. This is specially true for off-Internet countries like Pakistan where transit bandwidth is still holy spirited water. Since, for most providers, it is just a matter of one 2X2 sq ft data center space and the electricity bill (typically four mid-range servers and an Ethernet switch), it seems like a no-brainer deal.

Tariq Mustafa

http://twitter.com/#!/tariqmustafa/status/47964235455664128

 

[Fawad Niazi]

A black box in your system ... :)

On Tue, Apr 12, 2011 at 9:41 PM, Zarrar Khan <zarra...@gmail.com> wrote:

[Zarrar Hasham Khan]

Dear Tariq,

I agree with your comments. However, my question was more directed towards the scenario where the operator already has a cache present on the IP transit stream. Does it make sense to have two caches in series in terms of user experience? What are the incremental benefits that Google Cache gives on top of original cache that justifies adding another layer of delay to the customer.

We understand that we will not cache youtube on our original cache if we install Google Cache.

Regards,

Zarrar

[Tee Emm]

GCC is more about distributed content delivery and less about caching. Based on the DNS element they have in their configuration requirements, I'd imagine that the users will directly get the addresses of the local boxes instead of the local boxes trying to get in the way of an otherwise outbound traffic.

Will contribute my own assessment, NDA permitting.

Tariq Mustafa

Phone # | Twitter | Profile | Industry | Work | Work Place 

[Majid Farid]

The cache you have on your Gi interface is good for caching all the
content however the cache mechanism on that verses the Google cache
is different in a sense Google cache is more geared towards Youtube
and Google content. If from analysis of your current Gi traffic you
see a substantial youtube and google traffic then you should discuss
the reduction in that traffic with Google if you decide to put GCC in
your network.

The ROI should be simple to calculate if Google provides SLA regarding
the reduction in your current upstream bandwidth due to its content
been cached locally and which should be timely as well

Regards,

/Majid

On Apr 13, 11:12 am, Tee Emm <tariq.must...@gmail.com> wrote:
> GCC is more about distributed content delivery and less about caching. Based
> on the DNS element they have in their configuration requirements, I'd
> imagine that the users will directly get the addresses of the local boxes
> instead of the local boxes trying to get in the way of an otherwise outbound
> traffic.
>
> Will contribute my own assessment, NDA permitting.
>
> Tariq Mustafa

> Phone # <http://tinyurl.com/tmcell> | Twitter<http://twitter.com/tariqmustafa>|
> Profile <http://www.google.com/profiles/tariq.mustafa> |
> Industry<http://groups.google.com/group/telecom-grid-pakistan/topics?hl=en_US>|
> Work <http://multinet.com.pk/> | Work Place <http://goo.gl/maps/Pwbz>

>
> On Wed, Apr 13, 2011 at 12:04 PM, Zarrar Hasham Khan

> <zarrar.k...@gmail.com>wrote:

>
>
>
>
>
>
>
> >  Dear Tariq,
>
> > I agree with your comments. However, my question was more directed towards
> > the scenario where the operator already has a cache present on the IP
> > transit stream. Does it make sense to have two caches in series in terms of
> > user experience? What are the incremental benefits that Google Cache gives
> > on top of original cache that justifies adding another layer of delay to the
> > customer.
>
> > We understand that we will not cache youtube on our original cache if we
> > install Google Cache.
>
> > Regards,
>
> > Zarrar
>
> > On 4/13/11 7:58 AM, Tee Emm wrote:
>
> > I am not sure about improvement in the startup delay but the amount of
> > transit bandwidth it could save on popular contents could be one big plus to
> > having this set up in place. This is specially true for off-Internet
> > countries like Pakistan where transit bandwidth is still holy spirited
> > water. Since, for most providers, it is just a matter of one 2X2 sq ft data
> > center space and the electricity bill (typically four mid-range servers and
> > an Ethernet switch), it seems like a no-brainer deal.
>
> > Tariq Mustafa
>
> >  http://twitter.com/#!/tariqmustafa/status/47964235455664128
>