Virtual Test Environment Help

[Mike_182]

Right guys, my brain is frazzled and I haven't the foggiest how to accomplish this. Also not sure that this falls under 'PC and PC Gaming' - I don't think these are particularly 'personal' computers! Lol!

I have:

£20,000 budget
2x HP ProLiant DL380 G5s, with dual E5345 processors and 4GB RAM, each with 2x72GB 10k SAS disks, and 2x146GB 10k SAS disks.
1x HP MSA1000, with a second shelf, first shelf has 14x300GB U320 disks, second has 10x146GB U320 disks. MSA has both controllers, but this is irrelevant for a test environment.
2x HP ProLiant DL360 G5s, with single E5420 processors and 4GB RAM, each with 1x72GB 10k SAS disk (don't ask!).

I can't spend money on new hardware, and I don't need to spend the money on licencing as we've got MSDN for the MS stuff and we'd use ESXi or XenServer. They need to have 2 environments, and would like a 3rd, each with around 10 virtual servers. These virtual servers will comprise of a DC/Exchange server, 3 SQL servers, 4 application servers and another random one which I can't remember off the top of my head, and will also have 3-4 clients - this is per environment. They need to be able to retain several snapshots of each device, and it needs to have no connectivity with the live LAN other than through the host management network.

How the hell do I do it? If I could buy new hardware, it'd be a c3000 with 4xBL480s, with the E5450 processors and 16GB RAM in each, paired with a HP MSA2012i and 12x750GB disks. However I can't buy new hardware. I can buy disks and RAM, but my concern is the amount of work (and data) involved is way beyond 2x DL360, and is more than 2 DL380s of that spec can manage.

Help!

 

[dk]

not sure i completely understand the question, you say you have £20k budget but then that you can't spend money on hardware?

Then you say its a test environment, but you have 20k, thats an expensive test environment?

Surely all you need to do is install esxi on the DL's and then present the storage from the MSA1000 to it? You will need some HBA's but thats all, presume the MSA1000 has the built in 8 port switches?

To do the snapshots will be difficult as you need vcenter with vmware to take snapshots of machines, the networking is no problem as you just create your own networks and vlans using the esx vswitches.

The hardware is fine, those machines will run at least 10 virtual machines with those processors, just need some more ram, but with a 20k budget that should be no issue.

have I got the right end of the stick or are talking about something completely different? thos servers with some more ram will more than handle what you have mentioned, more than handle. Especially as its a test system so won't have many users connected, if you had a couple of hundred users connected it might be an issue but just shove 32gb ram in and you will be fine.

 

[Mike_182]

Yeah - you kinda see the problem I face. I can't use the MSA1000, that's apparently allocated for something else, and I can't buy 'new' hardware, any money has to be spent on upgrading old kit. The MSA1000 doesn't have the built-in switches, but fortunately the one thing I do have is 24 ports free (2x12) on a fully licenced HP SAN switch. But I can't use the MSA.

I think TBH they've got to suck up and buy a bit of new kit - whether it be an all-out 'sod it, for the £12k out of £20k for c3000 and 2000i it's a snip' or whether it's an MSA for the DL380s, it's gotta have a decent amount of storage - and that's one thing you can't manage in the front of a DL380.

Will be most likely swaying towards XenServer, as it does everything we need of it in the free version - and as much as I'd like to pay for VMware and have shiny toy with funky s**t, it's an unnecessary £10k expense for something that doesn't really need support - if it stops working, flat it and start again sort of thing!

But yeah - the £20k budget is:
a) excessive if using free VMware
b) restrictive if you want to use ESX and VMotion
c) not available for use on new tin

F**king annoying - cos for that money they could have a singing and dancing test environment which would cater for everything - instead, we're stuck with the b****cks old stuff. And I'd love to whack an extra shelf on our EVA, but that WILL be the whole £20k, and I've never known anyone so blaze with money to stick a test environment on an EVA!

Maybe when we get an XP, but even then...

 

[Mike_182]

Oh, and I forgot to mention that I can have either the two DL380s, or the two DL360s. Not both.

When I first got the request, I thought - great! Nice and simple, the hard bit is extracting from the business what they want it to do! Not that I'd be doing it with my hands tied behind my back...

 

[dk]

well you can get 8x300gb sas drives in the dl380, so its not that restictive (the g6 model take 16 drives!)

MSA1000, an extra shelf would be that much, the msa30 is about £1200 and the disks about £300 each so max £5k really, although the msa30 is ahrder toget hold of these days because its EOL now.

But if you aren't using things like vmotion you don't need shared storage anyway so pump the 380's full of 300gb disks, whack the ram up and bobs your uncle.

 

[ivegota16v]

you really really really badly need more memory to run that amount of vms, its going to kill you environment with that little! specially with what you want to do

ideally if you can get some cash replace the 5300 series with some new 5500 series, there peformance in vms in nearly 2-3x the performance and power usage of easily less than half, i have some white papers that show if you replaced what you have there with some new 5500 series you will have ROI in under a year! ! let alone the savings in space, cooling etc etc

worse case scenario... MORE RAM!!! it will work, but not to hot! lol

can you not get any new processors at all? or boards etc? u could in theory run your whole test environment on one quad socket 7400 dunnington system

GP

 

[dk]

you really really really badly need more memory to run that amount of vms, its going to kill you environment with that little! specially with what you want to do

ideally if you can get some cash replace the 5300 series with some new 5500 series, there peformance in vms in nearly 2-3x the performance and power usage of easily less than half, i have some white papers that show if you replaced what you have there with some new 5500 series you will have ROI in under a year! ! let alone the savings in space, cooling etc etc

worse case scenario... MORE RAM!!! it will work, but not to hot! lol

can you not get any new processors at all? or boards etc? u could in theory run your whole test environment on one quad socket 7400 dunnington system

GP

he can't get 5500 processors without buying G6 servers, the motherboard and architecture plus ram is TOTALLY different.

You can get up to twice as many vm's (in theory) on a g6 server running high end 5500 series processors but its not just the processors that are helping, a s**t load of more ram (the benchmarking has been done with 144gb ram, something the g5 could only dream of) QP from intel, better raid controllers etc etc.

So in summary, he'd have to buy G6 servers (which are more expensive than the G5 mainly because the ram is nearly twice the cost) and he can't buy new hardware.

These white papers you mention, have you read them properly, they get twice the vm's due to the processors as you say, but also the fact that they put 144gb of ram in the server, do you know how much that amount of ram would cost for a DL380, i'll tell you, a s**t load of money!

 

[ivegota16v]

he can't get 5500 processors without buying G6 servers, the motherboard and architecture plus ram is TOTALLY different.

You can get up to twice as many vm's (in theory) on a g6 server running high end 5500 series processors but its not just the processors that are helping, a s**t load of more ram (the benchmarking has been done with 144gb ram, something the g5 could only dream of) QP from intel, better raid controllers etc etc.

So in summary, he'd have to buy G6 servers (which are more expensive than the G5 mainly because the ram is nearly twice the cost) and he can't buy new hardware.

These white papers you mention, have you read them properly, they get twice the vm's due to the processors as you say, but also the fact that they put 144gb of ram in the server, do you know how much that amount of ram would cost for a DL380, i'll tell you, a s**t load of money!

course i do! i help write them! applogies if you thought i meant just drop the 5500 series part in!

i just had my 128gb of ddr3 delivered the other day so i do indeed know the cost!

here a quick roi demo http://www.youtube.com/watch?v=YJv7eKXR8yc not based around virtualization so to speak, but shows the ROI

its not just the ram though, its the EPT enhancments, the QPIS etc etc

but its also proven that on the 7400 series, a quad socket machine can run 2-3x the vms of the 5400, but again yes ram is a big part

in mikes case i think he is going to struggle with what he wants to do, he has the storage aspect easy but not enough grunt so to speak....
especially if he cant spend the cash on either new cpus or mem. a quick one Mike, how come you have 20k but cant spend it on new hardware?

 

[Mike_182]

a quick one Mike, how come you have 20k but cant spend it on new hardware?

Good question. One I don't know the answer to, other than the powers that be say so.

And as for ROI - if I've got a room full of G5s, which don't need replacing or beefing up, why do I have to invest? If I don't invest, there's no return on it, and let's be fair, the G6 (whilst a good bit of kit) isn't worth walking into your server room and pulling out all your G5s...

 

[Mike_182]

These white papers you mention, have you read them properly, they get twice the vm's due to the processors as you say, but also the fact that they put 144gb of ram in the server, do you know how much that amount of ram would cost for a DL380, i'll tell you, a s**t load of money!

That amount of RAM for a DL380 G5 costs 3 servers plus s**tloads...

 

[ivegota16v]

you do get a good 2-3x the perfomance on the G5 v the G6

theres going to be a video released that i did today in the lab showing an easy 2x increase in performance on the 5450 v the 5540 and using shite load less power

but yes, i do indeed see your point cant blame a guy for trying! got to get that stock price up! if you have a lab full of old 5100, or 5200 single cores then they do pay for themselves in a very short time !

 

[Cookie]

I f**king hate when people move goalposts on test projects

Edit: That video is pretty

 

[ivegota16v]

on a side note though, cant you work with a supplier who will put it through as an "upgrade" cough cough, when really its not if its all about capital etc

ie that new "ram" or "software" was very pricey

GP

 

[dk]

you do get a good 2-3x the perfomance on the G5 v the G6

theres going to be a video released that i did today in the lab showing an easy 2x increase in performance on the 5450 v the 5540 and using shite load less power

but yes, i do indeed see your point cant blame a guy for trying! got to get that stock price up! if you have a lab full of old 5100, or 5200 single cores then they do pay for themselves in a very short time !

as you seem to suggest you either work for intel or hp, i have a question for you.

This was something a customer asked me the other day and i couldn't answer.

Turbo boost, what OS's support this please, as with the feature it does state that the OS needs to support it, i'm guessing windows 2008 does but does 2003?

Also, this shared L3 cache, hows it better than the individual cache, the 5400's has 2 x 6mb cache, now they have 8mb between them. I can understand that as they are sharing its more efficient as the data held in cache only has to be written once but its smaller anyway as its 8mb compared to 12mb and whats wrong with them having their own cache?

I couldn't find anything that gave a finite answer as to how it was better in the real world except for marketing saying it has 8mb of L3 cache.....?

 

[Mike_182]

I f**king hate when people move goalposts on test projects

Never mind moving goalposts, these are doing f**king loop-the-loops.

 

[ivegota16v]

as you seem to suggest you either work for intel or hp, i have a question for you.

This was something a customer asked me the other day and i couldn't answer.

Turbo boost, what OS's support this please, as with the feature it does state that the OS needs to support it, i'm guessing windows 2008 does but does 2003?

Also, this shared L3 cache, hows it better than the individual cache, the 5400's has 2 x 6mb cache, now they have 8mb between them. I can understand that as they are sharing its more efficient as the data held in cache only has to be written once but its smaller anyway as its 8mb compared to 12mb and whats wrong with them having their own cache?

I couldn't find anything that gave a finite answer as to how it was better in the real world except for marketing saying it has 8mb of L3 cache.....?

ok, the turbo boost one..

Turbo boost works by the os requesting a new c state that the processor can go into, it is turned on and off in the bios and it will work with an acpi aware OS . so its supported in 2k8, 2003, redhat 5.1-5.3 (nehalem updates) etc etc

theres a good white paper here http://download.intel.com/design/processor/applnots/320354.pdf?iid=tech_tb+paper

basically if the processor has the TDP then it can turbo up all the cores by up to 3 bin (about 400mhz) parts (on the 5570 +) or 2 on the lower models

one thing to remember with turbo mode is that it is intelligent , so say you only have a software app that uses only 1 thread, it will c state the other cores to shut them off, hence getting a load of TDP head room and then turbo up the one core that the thread is using + 3

The advantage of the L3 cache configuration is that a single-threaded application has access to the full 8 MB of L3 cache. This was not possible on the previous generation, since their 12 MB L2 cache consisted of two halves, each located on one of the dual-core dies that made up the quad-core package.

Another advantage of the large L3 cache is that all four cores can work with a single set of data, rather than having to duplicate it across several caches; this saves space and allows more data to be kept in the cache. Exchanging data between cores also benefits from a major speed increase compared to the previous generation with two separate caches. plus with the new HT this is more apparant, its all to show that the sever platform is intelligent and can adapt to any workload, wether its extremly parrallell or single threaded

phew, i feel like im back in the office! and i work for Intel not HP

but not to go to off topic on Mikes thread if you have any other questions feel free to ping me, ill answer what im allowed to

 

[Cookie]

What do you do for Intel, out of interest?

Turbo boost reminds me of those buttons on the old 386 platforms where you could go from 33 to 40mhz

 

[dk]

ok, the turbo boost one..

Turbo boost works by the os requesting a new c state that the processor can go into, it is turned on and off in the bios and it will work with an acpi aware OS . so its supported in 2k8, 2003, redhat 5.1-5.3 (nehalem updates) etc etc

theres a good white paper here http://download.intel.com/design/processor/applnots/320354.pdf?iid=tech_tb+paper

basically if the processor has the TDP then it can turbo up all the cores by up to 3 bin (about 400mhz) parts (on the 5570 +) or 2 on the lower models

one thing to remember with turbo mode is that it is intelligent , so say you only have a software app that uses only 1 thread, it will c state the other cores to shut them off, hence getting a load of TDP head room and then turbo up the one core that the thread is using + 3

The advantage of the L3 cache configuration is that a single-threaded application has access to the full 8 MB of L3 cache. This was not possible on the previous generation, since their 12 MB L2 cache consisted of two halves, each located on one of the dual-core dies that made up the quad-core package.

Another advantage of the large L3 cache is that all four cores can work with a single set of data, rather than having to duplicate it across several caches; this saves space and allows more data to be kept in the cache. Exchanging data between cores also benefits from a major speed increase compared to the previous generation with two separate caches. plus with the new HT this is more apparant, its all to show that the sever platform is intelligent and can adapt to any workload, wether its extremly parrallell or single threaded

phew, i feel like im back in the office! and i work for Intel not HP

but not to go to off topic on Mikes thread if you have any other questions feel free to ping me, ill answer what im allowed to

thanks for the info, very interesting!