Reply to: Network advice needed

[Sysman]

BGG,

Thanks for the other side of the coin.

[Board Game Geek]

Just to jump in with my own experience of the HP Procurve kit.

We had 60+ sites, and each site had a 24 port 10/100 hub, mostly Netgear, but some Linksys and a few other brands I cannot even recall.

Some of these sites had perhaps as many as 22 PC's on a hub, and some had as few as 3.

We wanted to upgrade our network infrastructure, with the main purpose being to get switches so we could set SNMP on the devices and add them to a piece of SNMP monitoring software that we have back at HQ.

We didn't have the funding to go with fully Giga switches (just 2 ports per switch). Besides, most of the cabling in the 60+ sites is Cat 5.

We opted for the Procurve 2650 range, and they have achieved the objective. I can now troubleshoot the sites, from HQ if necessary, rather than have to drive to the site.

I get full stats on each port, can see collisions, and various other bits of info necessary to solving many network issues.

More importantly, I would estimate that in over 30 sites we have deployed these, the Procurve has flagged up Duplex Mismatch errors on our Internet Routing equipment which BT manage and put in (and we have no access to those Routers).

Armed with the error logs from the Switches, we have been able to present a solid argument to BT, and they have dialled in and fixed their Router. Net result has been a major increase in speed to and from the internet for the site.

Had we still be using hubs, we would have never known. Nor would the users at the sites, since they had go used to the slow speed and didn't tell us.

Out of the 60 sites, I've seen 3 Procurves replaced in 3 years, so not a bad failure rate.

Of course, other folks may have horror stories with the Procurves, and if so, then of course you'd be loathe to recommend them.

However, for our network size, the Procurve's have been ideal.

4 years ago, when the Procurves we're being ordered, I must admit I was a bit dubious and thought that we should go Cisco, like we do in the HQ Datacentre and BackBone, but for the smaller, remote sites, HP's have worked out ok.

YMMV

[Sysman]

I don't assume anything.

http://www.hp.com/hpinfo/newsroom/pr...8/080527a.html

"According to the analyst firm’s data, ProCurve, the world’s second largest enterprise LAN networking vendor, grew worldwide port shipments by 28.4 percent in the first calendar quarter of 2008 compared to the same period in 2007. The industry growth rate for this same period was 7.82 percent."

They are doing all-right, can't see them being that bad - value for money most likely, just the kind of purchase I personally like.

That stuff is marketing BS. They are simply saying that in 2008 "port shipments"* grew faster than the rest of the industry (wouldn't be hard if they had only recently started with the ProCurve brand), and agressive pricing obviously helps.

* a "shipment" does not necessarily mean a sale - see what I mean about marketing speak?

Edit:- quoting "port shipments" is misleading too

But I've seen HP completely stuff up major clients over recent years so I am biased. I wasn't surprised to see this lawsuit recently.

[snaw]

Sometimes you are better to strap the inter server / switch links to 1000mbs full duplex than allowing them to remain on auto. – file transfer test will prove.

Umm, 1GB can only be full duplex. You're thinking of duplex mismatches with 100MB links ... and a simple show on the switch will prove (It'll tell you what it's set at and if you see a tonne of FCS, alignment & runts on the port.

[wxman]

Back to the original question …

For 200 users I would def go 1Gb between servers, with 1Gb drop down to 100Mb switches that distribute out to the users etc.

I am not sure what the server requirements are for 200 users? But assuming that they are “normal” IE a couple of file / print servers, email server, SQL server etc then 100mb to the users will be fine.

*If* the network then appears slow then go back to basics – actually time how long a 1Gb file transfer takes between servers and then from the server to the user – try to find the bottle neck.

Sometimes you are better to strap the inter server / switch links to 1000mbs full duplex than allowing them to remain on auto. – file transfer test will prove.

Slow networks are often (blamed) but found not to be the bottleneck – IMO slow server disk subsystems are the culprit – again a file transfer test will help to diagnose.

Look for horrid disk fragmentation (Windows servers) as well as physical disk read / writes – there are some really slow SCSI U320 disk out there believe me.

Is the server running a disk array, if so is the RAID controller set up for write through or write back?

Run a performance bench mark http://www.passmark.com/ on the server disk read writes - are they as expected.NB C: drives on windows servers are deliberately slowed down for disk writes (protection against power fail), D Drive is not so expect better read / write.

If the servers can serve up! as expected then look again at the basics.

I just “fixed” a slow 300 user network by replacing the cables used to patch the servers running at 1Gb with certified CAT6 patch leads – but here there were lots of errors reported on the switches with often server network dropouts. I was the second contractor in after the first one failed to spot the dodgy leads.

[AtW]

It's been so nice to have you away. Do us all a favour and leave again soon.

Why won't you put me on ignore list then? I'll post as often as I deem necessary - you'll be the last person who'd I care to ask whether you like it or not.

[DiscoStu]

No, 25% saved on this would not be worth it - I'd go Cisco in this case.





Should be over 10 TB in 24 hours.

It's been so nice to have you away. Do us all a favour and leave again soon.

[AtW]

Every one of those others could very well give up on it at some point in the future, not going to see that happen with Cisco. That's kind of important to most companies, of say around 200 users. They're going to shell out a stack of cash for technology they expect to be around for 5/10 years - for a 200 user network you're talking about what - 4 2960 switches and a couple of 3750's aggregtion switches - £10k on hardware with a Cisco solution, even if you save 25% of that, is that really worth it?

No, 25% saved on this would not be worth it - I'd go Cisco in this case.

But hey, you'd know best, cause like you transferred 5 TB of data last night.

Should be over 10 TB in 24 hours.

[snaw]

I don't assume anything.

http://www.hp.com/hpinfo/newsroom/pr...8/080527a.html

"According to the analyst firm’s data, ProCurve, the world’s second largest enterprise LAN networking vendor, grew worldwide port shipments by 28.4 percent in the first calendar quarter of 2008 compared to the same period in 2007. The industry growth rate for this same period was 7.82 percent."

They are doing all-right, can't see them being that bad - value for money most likely, just the kind of purchase I personally like.



What exactly is wrong with them?

Cisco will still be around in 3 years, that's true - I would pay huge premium for this however, and in current climate alternatives should be actively thought - you can justify your earned money by offering less than obvious solution on the basis of what you know about products, even an idiot can "recommend" getting Cisco/Oracle/Sun - smart pro would know value for money alternatives.

Of course you're always right AtW. How silly of me to doubt you. I've only been doing this for the last 17 years ...

OK as a product, I thought it was pretty pants, I installed 6 procurves for a client, two of them were faulty (Dodgy ports) and needed replaced, not a great start (OK, maybe it was a glitch, but still left a bad impression). Wasn't massivly impressed by them.

They've got a very small market share in the switching sphere (Cisco has arond 70%, everyone else shares the rest - from memory it's around 10% for HP).

Every one of those others could very well give up on it at some point in the future, not going to see that happen with Cisco. That's kind of important to most companies, of say around 200 users. They're going to shell out a stack of cash for technology they expect to be around for 5/10 years - for a 200 user network you're talking about what - 4 2960 switches and a couple of 3750's aggregtion switches - £10k on hardware with a Cisco solution, even if you save 25% of that, is that really worth it?

If you're talking bigger, then I've anecdotally that they don't scale well. Their main selling points are cost and lifetime warranty (Though it's not replace, generally it's refurb I believe).

But hey, you'd know best, cause like you transferred 5 TB of data last night.

[AtW]

I'm a network guy, not a Cisco guy AtW - I'm vendor neutral, just because I'm CCIE doesn't mean I haven't spent many years working with other network kit (You're first assumption and mistake).

I don't assume anything.

http://www.hp.com/hpinfo/newsroom/pr...8/080527a.html

"According to the analyst firm’s data, ProCurve, the world’s second largest enterprise LAN networking vendor, grew worldwide port shipments by 28.4 percent in the first calendar quarter of 2008 compared to the same period in 2007. The industry growth rate for this same period was 7.82 percent."

They are doing all-right, can't see them being that bad - value for money most likely, just the kind of purchase I personally like.

My opinion, after actually using them - they're tulip

What exactly is wrong with them?

Cisco will still be around in 3 years, that's true - I would pay huge premium for this however, and in current climate alternatives should be actively thought - you can justify your earned money by offering less than obvious solution on the basis of what you know about products, even an idiot can "recommend" getting Cisco/Oracle/Sun - smart pro would know value for money alternatives.

[snaw]

http://www.networkworld.com/community/node/43427

Possible bonus - they probably don't require an expensive Cisco certified person who does not know what the difference between latency and bandwidth.

HTH



P.S. In the last 12 hours our fifty quid Gbit switch dealt with 5 TB of network data sent between servers.

So an opinion piece, by a guy writing about HP's pricing should be the basis for making a decision about network technology? Once again you;ve surpassed yourself.

My opinion, after actually using them - they're tulip. Cisco have some decent cometitors in many arenas, but the pro-curve isn't one of them.

I'm a network guy, not a Cisco guy AtW - I'm vendor neutral, just because I'm CCIE doesn't mean I haven't spent many years working with other network kit (You're first assumption and mistake).

Possible negative, to your possible bonus - in 3 years time I can say Cisco will still be around, unlike a lot of these cometitors, and that you're guaranteed to find someone with the skillset to manage them. And at the low end (Which a 200 user install is - inc your wee setup) they generally don't cost all that much.

As an aside, what does volume of data transferred have to do with anything? I can transfer 720Gb of traffic across the backplane of a Catalyst 6500 every second ... it's irrelevent to the design, unless you need that kind of speed and if you're doing it across a network (As opposed to in a switch) then it sounds like you need a SAN.

Really, you're occasionally incredibly ******* clueless. Telling us how much data you're transferring really is quite pathetic.

[AtW]

I was talking to one of our Cisco consultants and he told me that Windows cannot use more than 500Mb anyway. There is some limiting factor in the way the network stack is threaded (or something)

Bulltulip: I've copied data across network from disk to disk at around 65-70 MBytes/sec effective, the main reason for not going faster was disk write limit on the other end. The only reason I don't do it often is because it is faster to move disks and copy data locally in parallel.

[JoJoGabor]

I agree with Snaw, 100Mb to the desktop. I was talking to one of our Cisco consultants and he told me that Windows cannot use more than 500Mb anyway. There is some limiting factor in the way the network stack is threaded (or something)

[AtW]

AtW I would ignore, as per normal - HP Procurve switches are carp

http://www.networkworld.com/community/node/43427

Possible bonus - they probably don't require an expensive Cisco certified person who does not know what the difference between latency and bandwidth.

HTH



P.S. In the last 12 hours our fifty quid Gbit switch dealt with 5 TB of network data sent between servers.

[snaw]

Go 100MB to desktop, gig for servers (If needed) then etherchannel interswitch links to give 1, 2, 4, 8 GB etc.

Really depends on your traffic profile, office layout, budget etc.

Companies may avoid 10GB for many reasons, cost primarily - hardware, external lines etc. If you're shipping about lots of data then you may want to take a look at why that's happening.

GB to the desktop I'd avoid unless there's an obvious reason, and I've not seen one yet in any company I've worked at - users don't need that much bandwidth (Servers rarely do to be honest, though with the advent of virtualisation that's changing) and you're in danger of overloading your backbone if everyone is gig attached.

There is no right answer, really depends on your circumstances.

As for why you'd go Cisco, again cost is a factor - Cisco will cost you more, but as you grow (If you grow) then it's harder to get those other skills, or know the company is going to be around etc (Think Nortel). 200 users, I'd go that direction.

AtW I would ignore, as per normal - HP Procurve switches are cr4p (I know I've played with them), netscreen are decent, but that's all, Cisco bought netsys .. wow Cisco buying a company? Almost every technology they offer is acquired - what's your point, or is it just another moronic I know best outburst?