Server equipment

[MatthewB]

It's not really a question, but it's not a rant and it's a bit too tech related for WOT

I've almost secured board approval to virtualise our entire environment. Everything we have is ancient, out of warranty and generally clapped out. Here's the problem, they're a bit on the tight side so I may have to run with Dell Servers rather than my preferred HP in order to get the better spec hardware.

Does anyone have any positive experiences with the newer stuff? My first and last brush with Dell servers were the 2950 and 19xx (both series 8?) about 5 years ago. Not only were they badly built (flimsy, rattled and hot) but the support was horrific.

*edit* As a paying member I'd rather not have my posts auto-buggered by an ad bot. Nothing in the options to turn it off? Otherwise It'll be a crafty edit to avoid it.

[GhostyDog]

Originally Posted by MatthewB

*edit* As a paying member I'd rather not have my posts auto-buggered by an ad bot. Nothing in the options to turn it off? Otherwise It'll be a crafty edit to avoid it.

What is that all about?

Have you thought of Fujitsu Servers? I've had a couple at another business both very good and very economical.

[MatthewB]

Originally Posted by GhostyDog

What is that all about?.

It's added a hyperlink to D3ll in my post, not sure I like the thought of what I say being altered for the purposes of advertising. I've just edited this post using the swear filter avoiding tactic.

Originally Posted by GhostyDog

Have you thought of Fujitsu Servers? I've had a couple at another business both very good and very economical.

I hadn't actually, I'd already binned IBM as being expensive and by all account quite poor, but completely forgot that Fujitsu are still about.

Solaris 11 Sparc T4 ?

[MatthewB]

Originally Posted by Nev

Solaris 11 Sparc T4 ?

I'm not nearly talented enough to be playing with Unix. I am Gates' beeatch I'm afraid.

*I've just removed 20+ Netra T1 105's if you want them *

[GhostyDog]

Originally Posted by MatthewB

I'm not nearly talented enough to be playing with Unix. I am Gates' beeatch I'm afraid.

*I've just removed 20+ Netra T1 105's if you want them *

True techs live at run level 3

Seriously though check out Fujitsu, decent kit with enterprise class management bundled, I had no problems whatsoever with the boxes I put in an uncooled location in a satellite office.

[bobda]

We've used a few of the basic Dell R210's recently without issue. We don't quite use them as an IT install but they end up in Outside Broadcast vehicles running all sorts of SQL databases and control systems and so far without any issues. We like them because they're much more compact than an alternative HP (pretty much half as deep), they run cool and without excessive noise (except bootup) and they're around £700 a pop - excluding the RAM upgrade.

[NineOneSix]

E=MC2? Do sun microsystems still make servers or just software now?

Oracle bought Sun.

[NineOneSix]

Checked it out and seen that, seems they still do hardare though.

[JabawokJayUK]

I used to run a co-lo full of dells, 2950s etc and can confirm your original statements! My argument was always that they were the fastest hardware but it came at the csat of build and support! So cold spares were the only way forward!!!

HP is good quality these days and reasonable cost, but fujitsu do have some cracking kit! My days of running infrastructure are long gone but I built a vm based cloud last year, because I could, based on fujitsu hardware and it's very good gear.

My only suggestion is check the processor specs against the features you need. I fell foul of some older processors not supporting features in the os that I needed but the newer gear was just fine.

As for the *nix vs M$ debate....

Originally Posted by NineOneSix

Checked it out and seen that, seems they still do hardare though.

I think we'll see a resurgence in Sparc/Solaris with the new processors
and Solaris 11.

It's solid stuff. Runs for years without ever needing to be taken offline.
(All our terminal servers are still Sun boxes and I still run CDE by choice.
)

[GhostyDog]

Originally Posted by JabawokJayUK

HP is good quality these days and reasonable cost, but fujitsu do have some cracking kit! My days of running infrastructure are long gone but I built a vm based cloud last year, because I could, based on fujitsu hardware and it's very good gear.

My only suggestion is check the processor specs against the features you need. I fell foul of some older processors not supporting features in the os that I needed but the newer gear was just fine.

My data centre here is full of HPs 5th Gen kit and 'touch wood' no serious issues so far after 3 years, might even extend my lifecycle predictions as the kit seems very solid.

We bought a IBM xSeries server cheap off eBay (Quad CPU, loads of RAM, Brocade Fibre Interfaces x2, quad GB interfaces, 500GB RAID 5, £130) as a test rig and I really should have checked the spec more carefully as only the second gen of the server had CPUs which allowed for hardware virtualisation.

Ended up sticking an old copy of ESX on there instead of Hyper-V which I wanted to use for dev purposes.

Sickner, amazing Server for the price though, the Fibre interfaces were worth over a grand each originally.

[156Chris]

Originally Posted by Nev

I think we'll see a resurgence in Sparc/Solaris with the new processors
and Solaris 11.

Give that most IT is out sourced now ... how are they going to justify a massive support contract for them???

[JabawokJayUK]

Originally Posted by Nev

I think we'll see a resurgence in Sparc/Solaris with the new processors
and Solaris 11.

I have to disagree Nev, as per Chris's point, hardware has become a commodity game since virtualisation and general grid computing became mainstream. the cost of making Risc architecture chips (which I agree are way supior to x86 intel or AMD typically) makes the hardware unattractive. You only have to look at supercomputing to see that this old school arch is long since dead (I do miss my old Sparc desktop though . The latest supercomputers make use of GPU's for high speed calculations and AMD's for control which achieves way more for so much less, and lets face it, in the world of IT, more for less is a hard argument to ignore!

Sun gave up on sparc the second they released the x86 install of Solaris

[MatthewB]

Originally Posted by JabawokJayUK

My only suggestion is check the processor specs against the features you need. I fell foul of some older processors not supporting features in the os that I needed but the newer gear was just fine.

Going to be using current gen Xeon (E5649) x 2 per server. 24 cores (counting HT) mmmm

Originally Posted by Nev

I think we'll see a resurgence in Sparc/Solaris with the new processors
and Solaris 11.

It's solid stuff. Runs for years without ever needing to be taken offline.
(All our terminal servers are still Sun boxes and I still run CDE by choice.
)

Same with us, two racks full of T1's and V210's mostly running Solaris 9 that (according to my predecessor) hasn't missed a beat in nearly 10 years of operation only being taken down to upgrade from Solaris 8 It's almost a pity that it's in the process of being decommissioned.

[156Chris]

Another big downfall for Sun was them providing thin clients to run Windows, great in the short term but in the long term who is going to pay a premium for sun equipment just to run windows???


Microsoft have played it fantastically well with Active Directory and Group Policy, being able to control what users can do down to individual options being availble on individual applications is fantastic for control freaks sys admins ... but it means that it is nigh on impossible to have a mix of technologies on a single network that work together in a consistent fashion .. ergo everything has to run Windows (for microsoft)

[EddieGTA]

What HP would you have gone for? We by default were buying DL380 but recently virtualized our whole environment across a win 2008 r2 cluster across 4 DL 360 G7's. Storage is on fibre channel, but we are running everything include 3 exchange servers, 3 SQL plus file and print services.

Another plus is they are only 1RU, and can now take 8 internal drives.

[MatthewB]

Same as you, they would be DL360 G7's. I'm going to be using a SAN (if HP a P2000 G3), so providing they can take a couple of CPU's, a stack of memory and enough expansion for a few NIC's (I'm using iSCSI for the SAN) 1U is ideal.

As we're an MS Gold Partner we get quite a bit of MS Licensing so by default I'll be using Hyper-V (2K8R2 Datacentre) for each host and I'll be using SCVMM for management and DPM2010 for backup.

If we end up with Dell it'll be R610 servers and a Powervault 3220i SAN. Given how keen their pricing is for two servers, SAN (24x600Gb 10k SAS) and two switches I think we'll be having Dell.

[EddieGTA]

We're at 48GB and just ordered another 16 for each node

Hope you get the kit you want it was a steep learning curve for us but interesting to do.

[MatthewB]

I've specified 96Gb per server, a touch overkill perhaps I started this job with almost no virtualising experience, but a load of research, setting up a test environment with my own kit and reading "Mastering Hyper-V Deployment" by Aiden Finn has prepared me for what I'm about to face. I'm an IT team of one so there's no backup.

I do have the luxury of time, so I'll be building it, testing it, build it from scratch again having learned lessons testing it again followed by the P2V conversion of the Dev servers, more testing followed by the production servers

I still have to work out what to do about our Exchange server DAG and File Server DFS (both with another office in London) as they're not supposed to work on a Cluster Shared Volume.

[JabawokJayUK]

Originally Posted by MatthewB

I still have to work out what to do about our Exchange server DAG and File Server DFS (both with another office in London) as they're not supposed to work on a Cluster Shared Volume.

I heard that line from Micro$haft too... but they surprising do support it on the M$ virtualisation platform from what I could establish (don't quote me, check it out). I think its largely to do with disk IO.

On that note, make sure you have thought through your IO strategy, including server usage trends mapped to disk & hardware pools. I have a cloud/HyperV rig that my team use to do their job from (long story, cant go into details etc) and we have loads of ram, but the disk IO was wrong when I built it, so as soon as they get a few desktop instances up all the instances grind to a halt on IO performance.

[MatthewB]

I spent countless hours working out IOPS, in the end I worked out I'd need around 1200 average with a peak of 1800, this SAN should deliver 2400, so hopefully I've provisioned enough for the next 3 years. I'll also he using multiple NIC's and dedicated switches to make sure the iSCSI isn't the bottleneck either.

As for the DAG on a cluster, it is possible provided you don't make it highly available, which sort of defeats the object IMO.

[JabawokJayUK]

Sounds like you have it all sorted mate! good luck with the build!!!

Dont forget to sanitise and upload your finished diagram to Rate My Network Diagram so the world can revel in your masterpiece a few of mine are on there from a past life but these days I don't design or maintain, I just break instead way more fun !!

[GhostyDog]

Matt, doublecheck all those licences on your servers, anything oem you can't virtualise