PPRuNe resilience

[Feline]

Got to say that I'm pretty impressed with how resilient PPRuNe has become - last night there were more than 4500 concurrent visitors and the site was usable - and I would guess that the site is still running at well above its usual population --

Well done indeed to whoever is peddling even harder down at the coal face!

:D:D:D:D :ok:

[Saab Dastard]

last night there were more than 4500 concurrent visitors and the site was usable

You may be surprised to know that at its peak it reached over 11,000!!

Load-limiting was put in place to keep things running, as there was a short period when the database wasn't keeping up (dbase errors). Basically, priority was given to registered users over guests for accessing the forums. After the load-limiting was done the site was coping admirably with almost 9,500 concurrent users.

So yes, hats off to the folks at IB who are running the back-end servers and networks.

SD

[davidjohnson6]

I thought this was the very reason that web hosting should be outsourced to large server farms running the websites of lots of different companies ? Spreads the cost of the extra capacity needed for the occasional spike in demand amongst many websites, rather than just one having to bear all the cost. The same presumably applies to databases as well ?

[Jofm5]

Saab,

It's not that good and that is not coming as a criticism but as an observation.

When the AF incident hit the press PPrune was being hit in rumors and news by around 5k + people on the one forum and you could not view anything on the site for around 2 hours without hitting retry. a minority of the time it was a timeout in the browser - the other rest of the time it was a message from the site saying too busy.

From an educated guess your mysql database is limited by the number of connections and the concurrent connections was exceeded thus pages could not be delivered. The overwhelming of the site should not have been a problem with the people on board as even at its peak of a few thousand its really not that performent compared to many sites.


I thought this was the very reason that web hosting should be outsourced to large server farms running the websites of lots of different companies ? Spreads the cost of the extra capacity needed for the occasional spike in demand amongst many websites, rather than just one having to bear all the cost. The same presumably applies to databases as well ?


Its the very reason you dont go for hosting - you go for co-location, you put your own hardware in the the POP (Point of Presence) on the internet - if you go for hosting then you will be one of many on the same cluster typically.

If you have a single box co-located you will still hit problems similar - what is required is a dedicated cluster - easily done with a number of Linux type boxes running apache but the question is should the owners of pprune spend the money to cope with the peak around an incident or just provide enough to keep the active community sufficient - I think the answer is in the question.

Given unlimited resources the site would be highly avialable all the time.

[mixture]

Its the very reason you dont go for hosting - you go for co-location, you put your own hardware in the the POP (Point of Presence) on the internet - if you go for hosting then you will be one of many on the same cluster typically.

What a load of codswallop.

It's like saying "I fly a plane".

Yes there's hosting that's like a battered old flying school Cessna ..... but there's also hosting that's like a Boeing or Airbus with all the redundant systems etc.

Like most things in life you get what you pay for. Don't expect super duper hosting that can take the load if 11,000 people if you're only willing to pay the hosting provider 99p per month.

There's not much point going for co-location if you don't have the technical staffing to take care of it.

Any monkey can stuff a box or two in a rack in a co-location facility ...... it takes experienced engineers to get the most out of the independence that co-location gives you,

Given unlimited resources the site would be highly avialable all the time.

Ah yes, similar to the old...

Cost
Quality
Speed

Pick two..... debate


Let's be realistic here. PPrune is just a forum, yes it's a popular forum, but it is just a forum. It's not Amazon and therefore throwing money at it just to get five nines availability would seem a pointless activity in my opinion.

[Bushfiva]

My sites are hosted on virtual machines on physical clustered hardware. I pay for a certain spec of virtualized machine. When I need more power, I pay more and the virtual machine automagically turns into something more powerful. I don't mind that there are a bazillion virtual machines running on the underlying hardware, because I have a performance guarantee on the virtual devices. I can separate the processing from the bandwidth, which has helped me reduce costs tremendously. As well as doing my own backups at the file level, the entire virtual machine is snapshotted from time to time, "time to time" being good enough for my needs. If I wanted to pay a lot more, I could go for virtual failover to physical hardware in a different location. Virtualization works very well for me.

[Jofm5]

Quote:
Its the very reason you dont go for hosting - you go for co-location, you put your own hardware in the the POP (Point of Presence) on the internet - if you go for hosting then you will be one of many on the same cluster typically.
What a load of codswallop.


Oh sorry I do apologise to your much superior knowledge and bow down to your feet - I only work in the industry so I know nothing of course.

Now lets try get some things clear here.

1. Hosting

This is where your website is "hosted" on a web server along side others and is the cheaper option as the cost of the hardware, maintenance and upkeep is spread amongst the hosts for the "hoster" to be able to make money. This hardware is maintained by the hoster with a service level agreement and typically nowadays is virtualised across a number of servers with resources restricted by the virtulisation software (typically VMWare)

2. Dedicated Server

This is where by the company that provides hosting facilities provides a dedicated server(s) for the web site in question - whilst this is better than hosting in that there is dedicated equipment and resources there is the limitation that the servers fall within the configurations provided by the company in question. Typically there will be no virtualisation or sharing of resources it will be a standard build that is maintained by the company in question.

3 Co-Location

This is where you purchase the equipment, set it up how you wish and co-locate that equipment within the racks of the communication provider of your choice. Typically the co-location provider will have 24/7 online support for you to request re-boots etc of hardware. This equipment and its configuration is under your control so can be as flexible as required, i.e. it can be a single 1U box or your own server farm - whichever is most suitable for the situation.


Yes there's hosting that's like a battered old flying school Cessna ..... but there's also hosting that's like a Boeing or Airbus with all the redundant systems etc.


I neither fly an airbus/boeing or a cessna, I write software for a living - so that analogy is a bit lost on me. I will refer you to the above as to how "hosting" a web service is sold and you can fit your analogy around that if so suits.

So now we have established some facts - which part exactly is codswallop in what I said??????


--------------------------------------
Bushfiva:-

Virtulisation does work very well in the respects you say and I wont dispute that - it is something we use extensively. The ability for a failed virtual cluster to be re-started on a new server cluster in minutes is an invaluable tool for hosting a service. However my point remains that if you are hosting many busy services on the same cluster then your resources are limited/shared and that dedicating those resources is a way to improve your service.

[mixture]

Jofm5,

(1) I'm not interested in getting involved in willie waving arguments. Let's just say my knowledge of the IT industry is not as far removed from reality a you may like to think. The fact that I might have spent all of two minutes typing my post this morning (nor do I have any intention of spending any longer now) , and therefore not been as eloquent as I might have been is another matter. :cool:

(2)
This is where you purchase the equipment, set it up how you wish and co-locate that equipment within the racks of the communication provider of your choice. Typically the co-location provider will have 24/7 online support for you to request re-boots etc of hardware.

You have hit the nail on the head here.

The pure definition of co-location is that it's your kit and you fix it. Assistance from co-lo facility is limited to "intelligent hands".... i.e. button pushing.

As I said earlier ....

Any monkey can stuff a box or two in a rack in a co-location facility ...... it takes experienced engineers to get the most out of the independence that co-location gives you,


If you can afford the capex involved with hardware and licensing, both initial and lifecycle.... then great.

What really counts more though is the need for staffing in-house to do network (e.g. BGP) and server support at two in the morning when something goes wrong.

Lots of companies see it making more business sense to outsource to a hosting provider so they can stay tucked up in bed.

With a properly negotiated contract and SLA it can be at least as good, and in many cases, better than trying to do it all in-house.

There are lots of very capable hosting providers out there (e.g. Rackspace). Don't tarnish them all with the same brush.

Look at all the big companies that outsource the hosting of their DNS despite all the competence and resources they have in-house (e.g. Amazon use Ultradns for hosting DNS).

The same goes for companies that outsource email anti-spam and anti-virus filtering despite having all the competence and resources in-house.

Co-location can be done just as badly as hosting. It's not the magic answer to dealing with reliability or scalability issues.

What you have said to bushfiva is also questionable. It's all about the infrastructure and how you use it !


(3) Each is entitled to their own opinion. I think we'll have to agree to disagree. This is an aviation forum, not somewhere to argue complex IT topics ! :cool:

[Saab Dastard]

OK, mixture & Jofm5, you've both had your turn.

SD