Topologies/Switch Throughput

[SYNACK]

Synaesthesia what are you going to use for routing between the VLANS as you'll need a layer 3 device to route between the subnets. You may be able to use a server with a bunch of NICs but that is going to have huge latency compared to a dedicated layer 3 switch (not sure if you have one floating around there).

[synaesthesia]

Cheers CyberNerd.

 

Was setting up 3 switches today with a "core" (a 48 port 3com 4500) so I could experiment with Vlans so we know what we're talking about before we dive in!

Couldn't find the IP address so looked at DHCP leases on the server. Oddly enough we found a DHCP entry with no name. Pinged... interesting. Went to its web address. HP ProCurve 2626. Ugh! So where the hell is it? Half an hour later we tracked it down to serving a staffroom and a few more machines. 100MB switch. Can't have been helping so that's being pulled.

Another procurve failed too, but HP thankfully are damn good with swapping so a new one is on the way.

Think we'll be plumping for the HP V1910-48G to bolster the core - it certainly has to be an improvement!

[CyberNerd]

Cybernerd please keep your ludditrey to yourself, not everyone is happy running a ten year old OS with limitid support and lesser security. Windows has moved on and so have many users. This is not the topic of this thread either and distracts from the problem at hand (no its not Windows 7).

 

And keep your fanboism to yourself- XP has been updated regularly since its release 10 yrs ago, SP3 is barely 4yrs old, and there are free modern operating systems that don't drain money. It seems ridiculous spending money on pc's and operating systems without spending anything on an infrastructure to run it on.

[SYNACK]

And keep your fanboism to yourself- XP has been updated regularly since its release 10 yrs ago, SP3 is barely 4yrs old, and there are free modern operating systems that don't drain money. It seems ridiculous spending money on pc's and operating systems without spending anything on an infrastructure to run it on.

 

Look, lets take this to PM if we must and keep it out of the OPs thread. The primary code base is older than 4 years, its just had updates tacked on top rather than having things reimplimented you hate Windows (7) we get it. Linux vs Windows is a seporate matter all together and will just trash the rest of this thread which is unfair to the OP.

 

 

 

Synaesthesia - that V series switch does look to have limited layer 3 ability so should be able to route between VLANs. You'll want to slice up your network into subnets so that it can be routed as just putting the same client range accross multiple VLANs will break connectivity. The subnets are the structure that allow communication between sections.

 

For DHCP you can look into DHCP helper on the core switch which takes DHCP requests from the different segments and fowards them to your server on whatever segment it may be on.

[synaesthesia]

Cheers, I've also been looking at those features but have avoided playing with anything unnecessarily.

 

My quick tuppence on the XP vs 7 is a simple one. If there's any modern system (post Core2Duo era) that runs better on XP than on 7 then it's severely broken.

[Guest Guest]

HP did have some of its lowest end switches made by other verdors and rebranded before the 3com buyout so it is still a possibility

 

I can confirm that (atleast some of) the V series is simply rebadged 3com kit. We have a 5 port HP switch which is exactly the same as the 3com one we bought, both at around the time of the merger.

[CyberNerd]

My quick tuppence on the XP vs 7 is a simple one. If there's any modern system (post Core2Duo era) that runs better on XP than on 7 then it's severely broken.

 

The point I keep trying to make is that you shouldn't be running Core2Due era PC's using gigabit NIC's on the network that you have. The problem here is that older PC's won't be able to use their gigabit NIC's because the processor won't keep up. On faster machines you'll just end up flooding the uplinks, it will get worse as you put in faster and faster machines.

 

The primary code base is older than 4 years, its just had updates tacked on top rather than having things reimplimented you hate Windows (7) we get it.

 

Don't put words in my mouth. I just say it's a waste of money to buy an EES agreement and expect to run regular updates if synaesthesia can't afford a network. I'm suggesting ways to save money and do it properly.

If it comes down to money the infrastructure needs to be in place before the bells and whistles.

[SYNACK]

I can confirm that (atleast some of) the V series is simply rebadged 3com kit. We have a 5 port HP switch which is exactly the same as the 3com one we bought, both at around the time of the merger.

 

Yes, that one is rebadged 3Com, I was reffering to the bunch of Procurves that he had issues with before, if they were from the bottom segment then even before hp's purchase of 3Com I am pretty sure that some of those were made by different vendors and rebadged.

 

The 3Com stuff should be alright, I was not to happy about how the company itself handled itself but their hardware was usually fine and now they have corporate oversite from a different company. There again hp is not exactly run by champions at the moment (WebOS saga) but hopefully with the sacking of yet another CEO they will get back on track. Even with all this their network gear has always seemed really solid and well featured which is why we stick with them.

[SYNACK]

The point I keep trying to make is that you shouldn't be running Core2Due era PC's using gigabit NIC's on the network that you have. The problem here is that older PC's won't be able to use their gigabit NIC's because the processor won't keep up. On faster machines you'll just end up flooding the uplinks, it will get worse as you put in faster and faster machines.

 

 

 

Don't put words in my mouth. I just say it's a waste of money to buy an EES agreement and expect to run regular updates if synaesthesia can't afford a network. I'm suggesting ways to save money and do it properly.

If it comes down to money the infrastructure needs to be in place before the bells and whistles.

 

Horse leavings, 'you should not be running gigabit as you won't get the whole gigabit out of it', that may be so but you still get at least 300mbit/s out of it, a three times speed increase for $10 more hardware. Also what about things like imaging over multicast where the other links are not as much of a concern, taking one sixth the time to image something has got to be worth something. Also you should at least have the ability to step up to the higher speeds if required, I get so sick of some people who insist on getting 10/100 gear as 'its all you'll ever need/can use' then being subjected to the slowdowns that it causes. Why must some people avoid new technology at all costs no matter what the benifit.

 

As to upgrades, at a certain point it is no longer even worth having the infrastructure if your network is that old and cack that nonone wants to use it or needs a training manual from musem archives to be able to understand it. User experience and familiarity matters to but we're drifting off topic again.

 

The network loading caused by gigabit to the edge all depends on the usage patterns of the systems and the users, in many situations those pipes are not saturated as things get done quicker so that load disapates quicker too. There may be issues in certain deployments but that is something that can be dealt with with QoS and further tweaks/upgrades later. 100mbit to the desktop is not the answer, it is deffereing the problem and having gig ability gives you that much more flexibility in how you distribute the bandwidth if you need to.

Edited November 24, 2011 by SYNACK

[ChrisH]

What are the hp port fast commands does anyone have an example? All solutions on google always end up talking about the cisco commands.

[synaesthesia]

As was said, lets keep any bashing of people's ideas elsewhere. I'm all for constructive criticism and I think enough of us are grown up enough to agree to disagree on some things.

 

Sorry Synack, I think I missed one of your posts/replies. I'm not sure about routing between subnets, however having spent an hour discussing the same subject last night with one of the most genius networking guys I know (who heads up various technical teams at Cisco UK) we had two options: we have a cracking Cisco router provided by ISP but we're not allowed to touch it. On the other hand, any vaguely modern PC with a decent linux distribution and IP tables will outperform just about any layer 3 device under 2 grand. That's not alot, discounting layer 2+ stuff, but plenty (apparently) for a network of our size. He also mentioned STP can be a prime suspect for throttling ports too much - however we only enabled it after the problems started.

 

Didn't know about using STP on uplinks and RSTP on clients - what's the reasoning behind this? All the sales bullpoo just shouts "RSTP is better, thanks" with no actual information, as a result RSTP has been enabled across the board, on the units that support it (which is just the Dlink 1210-24s, not the 1224's or Procurves.) Easily done though, we have per-port control of the units.

 

We'll see how things transpire with the improvements over the last few days and then start seriously planning for the near future.

 

Many thanks to everyone that's chipped in - even the off topic stuff has been extremely enlightening and may hopefully be of use to others.

Edited November 24, 2011 by synaesthesia

[SYNACK]

What are the hp port fast commands does anyone have an example? All solutions on google always end up talking about the cisco commands.

 

(config)# span protocol-version rstp

HP: E-Series networking solutions - HP Procurve Spanning tree config?

[SYNACK]

RSTP is just rapid STP and designed for clients so it only runs a few tests before enabling the port and allowing actual network connection. Full STP whould be on any interswitch links which runs more through tests but takes about 30s for the port to become 'live' for network connectivity.

 

The linux PC route is valid but becomes a massive bottleneck (everything pushed through 1GB NICs) rather than running on a switching fabric. There is also a bunch more latency thanks to the larger software stack inbetween and the fact that it is implemented entirely in software. Still a possibility but better to tie it in to a device like a switch if that is an option.

[ChrisH]

HP: E-Series networking solutions - HP Procurve Spanning tree config?

 

Yes but given my configuration of my switches typically has 2 or 3 VLANS on it, Usually Building VLAN, VOICE, Printers etc which are all usually untagged (apart from Voice) and then I have my uplinks which will be tagged in all 3 VLANS. How do I define the ports that use STP and the ones that use RTSP. The configurations seem to apply to a whole VLAN and I haven't seen a particular example of it applying to a subset of ports. In a psedo code type of example I would expect to see:

 

Vlan 6 "MainBuilding"

Untagged 1-48 Mode RSTP

Tagged 49-50 mode STP

 

or am I missing something here and it just magically happens as it knows the difference between and inter switch link and an end user device?

[SYNACK]

or am I missing something here and it just magically happens as it knows the difference between and inter switch link and an end user device?

 

No magic, you need to set it up manually and in some scenarios you may need to use per VLAN STP or MSTP

 

http://www.hp.com/rnd/support/manuals/pdf/release_06628_07110/Bk2_Ch5_STP.pdf

Edited November 24, 2011 by SYNACK

[CyberNerd]

Horse leavings, 'you should not be running gigabit as you won't get the whole gigabit out of it', that may be so but you still get at least 300mbit/s out of it, a three times speed increase for $10 more hardware. Also what about things like imaging over multicast where the other links are not as much of a concern, taking one sixth the time to image something has got to be worth something. Also you should at least have the ability to step up to the higher speeds if required, I get so sick of some people who insist on getting 10/100 gear as 'its all you'll ever need/can use' then being subjected to the slowdowns that it causes. Why must some people avoid new technology at all costs no matter what the benifit.

 

My rationale here is that working with the lowest common denominator ensures stability. If you give machines 1Gb/s they have the possibility of using it; this can cause issues with older networks, inadequate switches and with mixed environments that have 10/100Mb/s and 1000mb/s devices. suppose 50 machines are sharing a 1GB uplink, any slowdown caused by inadequate switches dropping packets will cause TCP acknowledgements to be delayed/missed and this effect can slow down a network massively as the transmission rate is slowed. Now imagine how left out a 100mb/s machine would feel in a mixed 1gb/s environment with slow uplinks!

 

Simulation TCP 1: Reliable Transfer

 

Clearly I'm not against high network speeds! just that the network needs to be able to cope with it regularly and reliably for it to work correctly - I just think it's better to have all running correctly and without errors. If the switches and uplinks won't be able to deal with the traffic it makes more sense to slow the traffic. Think along the lines of running country lanes with min speed limits, you get more traffic through at 30mph than 100mph.

 

What are the hp port fast commands does anyone have an example? All solutions on google always end up talking about the cisco commands.

 

On 3com (I understand they use 3com OS on even new HP now, certainly some of our HP branded stuff has 3com ios)

it would be

 stp mode rstp

Edited November 24, 2011 by CyberNerd

[ChrisH]

Looking at a few of my switch configs they are set to RTSP already. I think some of the commands I was grasping for are in here:

http://h40060.www4.hp.com/procurve/uk/en/pdfs/application-notes/How_to_improve_and_harden_spanning-tree_configuration_Configuration_note_Dec_08_A4.pdf

 

The ones about the admin edge modes.

 

It looks like only my Procurve 2610 and 5610 know about these though. My 2650 and 2826 don't seem to have this and most of of my edge switches are 2650 so I guess I better dig out the specific manuals.

[synaesthesia]

Not going to say much until Monday but I think problems have been solved

 

Dlink switches: not a problem. In fact, they've been holding up rather admirably considering when we recreated the problem with the Procurves they fell over and died immediately.

 

So it seems the networking issues in general were not related to some machines only connecting at 100mbit. We discounted, wrongly, the computers as the culprit - the drivers for the nForce 10/100/1000Mbit ethernet were nackered. Replaced them with the "wrong" driver, a little hacking of the .inf file, streamed into the CC3 build area and job was a goodun. So that effected our DT department, library and a few other machines which will all get a rebuild during a certain quiet period next week

 

So the speed issues?

Kept pinging suspect areas and narrowed it down to one. Core > edge 1 > edge 2 > edge 3 (horrible daisy chain but no choice!) NM found a slight loop - someone plugged network printer directly into PC then wall port to wall port. Joy. That was removed anyway, a bit of jiggery with those edge switches, lo and behold pings started going through reliably and not dropping out ever few seconds.

 

Phone call from our techy at the other site. "You know that problem I was having with my PC? It's stopped. It's perfect"

 

Woot!

 

Won't say too much, I'm touching wood as I type this, until Monday so we can confirm. We will still be buying in better switches (L2+) for the core and will certainly be vlanning for the new network next year, it's pointless wasting time on it now.

 

We're really rather happy Should have a new procurve on Monday thanks to warranty but we've found another dodgy one today - and it's only a few months old. I hate them.

[synaesthesia]

Just taken delivery of a pair of Procurve 5406zl 44+4SFP POE switches at a ridiculously low price

Very happy! 2 weeks of testing ahoy!

[gshaw]

Haven't seen much of this thread apart from the last couple of pages but I'd stick with the HP 5400 series (zl) for your core and built the edge up with the E-Series switches. We've got the v1910s for our iSCSI network and although they work OK the ProCurve E series seem a bit more solid and easier to manage in the long-term. Also add up the lifetime vs 3 year warranty (and subsequent care packs) to get a comparative price.

 

As for speeds, no point having 1Gb to the desktop if it's wobbly and unmanageable.. buy good kit buy once

[synaesthesia]

Well, the cheap gigabit switches which are oddly enough holding up better than the not-cheap procurve edge kit are holding out just fine Either way, 3 grand for 30 switches - end max speed to desktop wasn't important, it was a step up from crap 3com baseline switches which had a bucketload of inherent issues.

 

As said, we've invested in those 2 cores. I would appreciate a little more advice if possible on vlans.

 

2 sites. A and B.

Each site has a shiny new ProCurve 5406zl. Connected via fibre (which will eventually be link aggregated - I won't say trunked because that now confuses the hell out of me as I've been reading about vlans and trunks which have nothing to do with them)

We need to stay simple. This is important. Any advanced stuff can wait until the new network when we will probably vlan off printers, wireless APs etc as well.

 

So, 2 sites. There are servers on both sites which will connect to each core. We wish to simply have Vlan1 for site A and Vlan2 for site B which would effectively kill off the unnecessary broadcast traffic over that fibre link (the trunk)

The site workstations and printers do not need to communicate with eachother, just the servers at each site. Internet router which is untouchable is at site B. Obviously site A also needs internet.

 

So far my understanding is that I would need to tag the ports on the core and each edge switch at site A as VID1 and site B as VID2 and create a bridge (trunk?) between them.

 

If the above is correct, do we need to separate DHCP via AD for each physical site? Each time I see about DHCP something crops up about DHCP helpers and the procurve documentation seems extremely vague on this.

Am I massively oversimplifying this?

[gshaw]

You have to be very careful when talking between people with Cisco-speak networking terms and HP-speak as the words "trunk" and "tag" don't mean the same thing between brands

 

If you set a Trunk port on ProCurve that's link aggregation aka EtherChannel in Cisco speak

If you tag ports to a VLAN (as you say for link between two switches) that's a trunk in Cisco world

 

See Trunking Between Cisco and HP Switches « Andrew Travis's Blog

Edited December 6, 2011 by gshaw

[CyberNerd]

luckily much of the newer HP uses the 3com based OS, and a link aggregation is just that.

[synaesthesia]

Indeed - I just try to be clear between the two whilst looking for advice and finding people talking about different things for the same words

 

Righto, a little diagram to help further explain.

 

http://i236.photobucket.com/albums/ff271/krytenofsmeg/BasicSetup.jpg

Edited December 6, 2011 by synaesthesia

[apeman]

You have to be very careful when talking between people with Cisco-speak networking terms and HP-speak as the words "trunk" and "tag" don't mean the same thing between brands

 

If you set a Trunk port on ProCurve that's link aggregation aka EtherChannel in Cisco speak

If you tag ports to a VLAN (as you say for link between two switches) that's a trunk in Cisco world

 

See Trunking Between Cisco and HP Switches « Andrew Travis's Blog

 

 

HP Trunks are NOT the same as "link aggregation aka EtherChannel in Cisco speak". HP Trunks are a dedicated HP protocol that performs a very similar jobs as Link aggregation but with a few extra features, HP Trunks only work between HP switches that support them.

If you need to connected multiple links between switches of different manufactures then you need to configure the link as LACP (Link Aggregation Control Protocol IEEE 802.3ad) at the HP end and the other switch end