Load Balancing

I know this one has been done to death in the past, but...

I'm curious to know what exactly Zen are "load balancing". Is it customer connections/sessions across the centrals, or what?

It is customer connections/sessions across the centrals.

OK, we're talking centrals, right - as in BT centrals? Not "gateways"?

*BUMP*

I'm not trying to be difficult or anything, I'd really like to know whether the use of the term "centrals" by Zen staff when they talk about load balancing means the BT Centrals or their (Zen's) gateways.

See this thread as well which suggests that ISPs cannot load balance centrals, only BT can do it.

I just want to be clear about whether the terms are being used correctly (or at least as I would understand them).

"I'm not trying to be difficult or anything, I'd really like to know whether the use of the term "centrals" by Zen staff when they talk about load balancing means the BT Centrals or their (Zen's) gateways."

They're the same thing. Load balancing is typically done by disconnecting a users PPP session on the home gateway which forces them to reconnect. When they reconnect they can do so to any of the ISPs gateways (providing they all accept the same realm). Simply disconnecting a bunch of users from a heavily utilised gateway spreads the load.

"See this thread as well which suggests that ISPs cannot load balance centrals, only BT can do it."

Smaller ISPs tend to have BT managed home gateways, typically Cisco 7200VXRs. Zen have their own gateways manufactured by Redback.

TT

Cheers.

I guess then in the thread I linked to previously, the others who have been discussing this would be talking about ISPs using BT managed home gateways.

Sorry Babylon 5, think i must have missed your follow up question.

TT is correct, load balancing is done by dropping users connections, they naturally reconnect across different centrals which spreads the load more evenly.

Just to be accurate, it is centrals that we kick from. Each gateway can (and does) have more than one central.

Sorry to keep on at this, but in this post it's claimed that Zen are NOT performing load balancing at all but are simply (and I quote) "...relying on the numbers game...".

So who is correct?

Its all just semantics really though, in the end it balances the load across the platform and I think it would be fair to say that a lot of ISPs refer to it this way. Calling it "Disconnecting users to evenly distribute load among the centrals" is a bit long winded and IMHO isn't as easily understandable as calling it load balancing.

Unfortunately BT do not use load balancers and users are put onto a central in round robin fashion (although I think that technically this is a very primitive form of load balancing), if you get a higher proportion of high usage end users on a single central that one could become congested while the others are at 10% load (this is an extreme example just to illustrate the point).

Also while it may not be the networking definitiion of load balancing it is definitely fair to say we are balancing the load across the platform, trying to get the utilisation of each pipe so that it is approximately equally low; as I began with it is just semantics really.

Edited by deleted (Tue 12-Jun-07 19:39:27)

Zen are forcing the load to become balanced. I just posted to the other thread, but essentially by disconnecting users (who then reconnect) Zen force the BT system to redistribute the load. BT *must* have some way of distributing the load as an ISP can have multiple pipes and gateways.

Therefore as you know that the BT system will spread the load, you disconnect users, knowing that the users you disconnect will then be 'evenly spread' over the available pipes when they reconnect.

So, say you have 2 pipes and 1 is at 75% capacity and one at 25%. To balance (assuming simple round robin balancing) you kick 50% of the users on pipe 1. That reduces both pipes to 25%, but with a whole bunch of users about to reconnect. Those users will be evenly spread across both pipes (in this example bringing both pipes to 50%).

Of course, it gets more complicated with more pipes and gateways, but it *is* load balancing.

Ruz

"So, say you have 2 pipes and 1 is at 75% capacity and one at 25%. To balance (assuming simple round robin balancing) you kick 50% of the users on pipe 1."

66%.

TT

Ooops yes

It was late. Hopefully the example works though

Thanks,
Ruz

http://status.zensupport.co.uk/index.php?serviceid=5&incidentid=595

Just been shunted from speedy Gay to sluggish Gauss!

Edited by ARD (Thu 14-Jun-07 13:23:07)

I always find gadamer the worse gateway for me. The rest all seem pretty much the same.

Blackmesa8

And again today:

http://status.zensupport.co.uk/index.php?serviceid=5&incidentid=606

Edited by ARD (Thu 28-Jun-07 15:48:40)

Why is this done so often?

To help balance the load!

Seriously, think of it as two parallel roads that both run to the same destination. One road gets busier and busier whist the other stays relatively empty. On the empty road you can do 100Mph, but on the busy road you can only do 5Mph - Big difference & I'd know which one I'd prefer to be on.

Now, if you were to balance this busy road then half the traffic is taken off the busy road and transferred onto the not so busy one, now both roads have an equal amount of traffic on and can both happily run at 100Mph.

Convert those cars to IP Packets & the roads into the various backbones of the internet and you'll see why load balancing is important.

Good network admins will always make sure that the loads on their servers is as equal as possible to keep the traffic flowing as smoothly as possible.

I'm not sure enforcing mass-disconnections is a great way of dealing with the problem. And I cannot believe this is the only way to achieve what needs to be done. It seems very sloppy especially when the solution impacts end-users equipment requiring unplugging of devices for at least 20 mins to resync properly.

I'm not impressed at all.

In reply to:

---

It seems very sloppy especially when the solution impacts end-users equipment requiring unplugging of devices for at least 20 mins to resync properly.

---

Have you ever had to do this?

In reply to:

---

Have you ever had to do this?

---

Several of the forced disconnects have shown up in my router logs and I've not had a stale session yet.

*touches wood*

i appreciate you may mean resync as in "work" however load balancing should *not* cause you to lose sync - it will just cause you to lose PPP which is very quickly re-established by all but the rubbishest of routers. some zyxels i've used will drop sync every time ppp is dropped, but have no bother picking it up again. other zyxels will only drop sync if ppp is dropped a number of times. my netgear dg834 does not drop sync no matter how many times ppp is dropped.

hth.

edit: decided to replace a word with rubbishest rather that something else that bypassed the censor!

Edited by deleted (Fri 29-Jun-07 00:59:33)

Yes I have on a number of occasions.

The connection is rock solid and always reliable. But a business class ISP forcing mass disconnections on customers does not impress me on little bit.

My guess is this the cheapest way to accomplish the load balancing they desire which hoping it impacts only a minority of users.

I have a Speedtouch 510 that is prone to having trouble with this occassionally. I wouldn't describe it as rubbish piece of kit. I would call forcing mass disconnects on users and praying users hardware can handle the reconnection seamlessly is a total rubbish way of managing things.

Edited by deleted (Fri 29-Jun-07 18:13:26)

Re :But a business class ISP forcing mass disconnections on customers does not impress me on little bit

And how many customers that would qualify as a 'business' are active in the wee sma' hoors {as a Caledonian colleague of mine was fond of saying}, when load balancing occurs?

Edited by deleted (Fri 29-Jun-07 18:33:37)

I don't know. But my office server is backed up over my Zen connection in the wee sma' hoors every night. And large file transfers are done in the wee sma' hoors. Good time for a business to do things without loading the connection during office hours.

Unfortunately it is the only way of ensuring a balanced load. BT were looking to introduce "session steering" sometime ago - whereby an ISP could choose which Central customers connect on during authentication - but they abandoned that, so we're left with a "road-robin" situation when customers connect. This means chance and BT outages can result in a mis-balance which needs to be rectified. It's far from the best solution, but it's all that's available.

kind regards,
Phil.

As SkyFire already pointed out, this is the *only* way. The alternative is massive pings and slow downloads resulting from overloaded centrals.

I agree, it's annoying if PPP drops cause a problem - but it's the only solution to a problem that's much more annoying (slow speed, rubbish pings).

In reply to:

---

but it's the only solution to a problem that's much more annoying (slow speed, rubbish pings).

---

Indeed! - I'd much rather a connection that "May" disconnect gracefully* once or even twice a week, than suffer slow congested speeds which means I've got to stay online longer than I really need to, I do have a life!

ADSL Max isn't really the product to be on if you want a 24/7 always on connection anyhow, you'd have a fixed speed product or have a more specialised setup for reliability and redundancy. Even if you've paid extra for an Office Max connection or the 'cheaper' Home Max connection then you'll suffer the odd resync with the exchange in any case when certain conditions occur, so the 'Odd' PPP disconection isn't bothersome especially if your router immedietely reconnects.

For the first time in two months of being with Zen, I witness my first disconnections the other night due to "load balancing", so it hasn't been a problem at all for me, and neither will it be. If I let little things like that bother me then I'd be a nervous wreak!

*Gracefull disconnection = only PPP down, not the full resync at the exchange.

Are there any settings you could change (reconnect automatically?) that might help...?

Load balancing by disconnections is a method used by quite a few ISPs and it shouldn't be causing the problems you describe.

Last time they did it I had to reconnect. This time no problems and it seems faster.