We recently suffered a total server meltdown, with Active Directory getting upset and crapping out exchange taking out our whole system. It's all "back" to how it was, but after near a week of outage, there's quite a few unimpressed people airing concerns that little, or no, disaster planning, redundancy or contingency has been allowed for. I posted a little on this recently (pre-crash) about the whole offline storage (not that would have solved this problem) and didn't really follow up on that thread because, well the solutions all came down to cost... and surprise surprise, those costs were deemed "too much" - but now in the light of nearly a lost week's worth of billable time, that money doesn't seem so terrible.

As such, we're having a serious pow-wow with our (external) IT people on Friday, and I've already a significant list of "must consider" items to discuss.... but one I wanted to check up on, before wading in and talking like I know what I'm talking about (when I really don't) is our actual network infrastrcucture.

At present, we have a collection of unmanaged switches acting as hubs for each of our islands of workstations (4-8 machines a piece) that then feed 5 runs to the server room (I'm told there's a switch there that takes these then runs to the server).

Now in my last office it was all workstation>managed switch>server - single runs apiece, and for all the other issues we had with that setup, our gigabit speeds actually felt like gigabit speeds - and my downloads weren't affected (or at least didn't appear to be) when the person sitting next to me was streaming a video - whereas at my current place, everything just grinds to a halt (network wise) at "peak" times (morning logins, lunchtime, end of day uploads, etc) and I'm convinced it's the ghetto-esque network tree of multiple (different unit) switches all chained together and trying to funnel their way in....

Am I right to assume that? I have regularly asked IT "what's the traffic like?" on the network when things get chuggy, and they never give a convincing response. We have no shaping, no throttling, no bandwidth limiting/boosting, or site-access restrictons (mostly because we're one of those "nice" places that "trust" people, that doesn't "believe" in controlling such things ) - but since it is (in my mind) directly affecting productivity (which, although not the be all and end all of our operations, is a concern) I just wanted to ask here first if it's worth raising it as serious objection/proposal for improvement.

Sadly, 'tis generally the way.


There are various bits of 'network discovery' software which will automatically survey the network and draw you a diagram of what is connected to what (I think you mean 'topology' not 'typology' by the way ).Sadly all the ones I know of are 'enterprise grade' and therefore cost serious money but maybe someone else knows of an OSS/cheap one? The first step to sorting the network is getting a proper idea of what it currently is, as I believe you've decide already. Being able to say to the management 'if THIS switch breaks, everything goes pear shaped' is a good tactic to getting some redundancy in there. Network reliability is always something worth being concerned about. Doesn't matter if a server is still running if nobody can see it...

Which resources are slow at lunch time? Is it just internet or are file shares slow too?
With the morning logons, is it just slow for computers to start up, or is the network slow for those who have logged in already?
How many servers, desktops and switches do you have?
And how confident are you that anyone knows exactly how those switches are wired up?

You might well find that slow internal switching is just a matter of ill judged uplinks between switches, with some old 10/100 box being in line between a more modern switch and the server.

Virtualising the server should be a primary concern, along with a BDC (Back-up Domain Controller). If the hardware barfs, you can just load the server onto a backup box running VMWare or Microsoft Hyper-V and be up and running in "minutes".

The replacement hardware doesn't need to be full strength for emergencies, it would allow you to work at restricted speed. Obviously having identical backup servers would be better, but is more expensive.

We are currently expanding our system.

The first thing I did, 18 months ago, was to throw out the mish-mash of different switches with different speeds from different manufacturers. We always had performance problems, with certain segments being slow - and we were using mainly windows terminals, connected to a terminal server, so the throughput wasn't that high!

4 HP ProCurve gigabit switches and we haven't had any problems since.

We are now looking at a pair of Dell servers with their intelligent SANs on dedicated 10gbps backbone, with one in a building down the street for replication, in case of disaster.

We are looking at VMWare with regular snapshots (as opposed to mirroring). The advantage of the snapshots is that we will lose around 10 minutes of saved work, if the primary server dies, but if people delete files by mistake or something gets corrupted, configuration errors etc. you can go back to the last working snapshot and carry on from there.

It isn't cheap. The Dell EqualLogic SANs come in at around 12K, I think.

Try talking to Dell directly. They came and analysed our network, looked at the throughput and came up with suggestions for the server hardware, switches and SANs. The analysis itself was free - they hope that, by proving how badly configured your network is, they will get business for the kit to get it running smoothly.

I presume HP, Fujitsu and the other big name server brands will do similar deals.

What was interesting is that our normal supplier, who uses re-branded Intel hardware, couldn't match the EqualLogic SANs, but the less capable kit he offered came in around 15% more expensive than the Dell offer, and the Dell offer included a 3 day workshop on configuring the network and 24/7 4 hour service.

Sounds like you need a complete overhaul of your setup.

Drop me over spec's etc via email and I'll put together some recommendations for you to talk to the IT guys about.