Recommended rack-mount servers?

[The Griffalo]

mystomachehurts said:

The Griffalo said:

The G5's seem to be noticeably quieter than the earlier ones but even so I wouldn't call them quiet still.

We bought a few G5s recently and I can't hear my mate over the phone in the same room when they start up.
Still, he says they settle down a bit once they are running.

Anyhoo Griffalo haven't you got any real work to be doing rather than arsing about on here?

(And no, I haven't got an electric fence I can go play with!)

Hmm, let me think about that.... Write PHP scripts or arse about on PH.... Tough call

[mystomachehurts]

The Griffalo said:

Hmm, let me think about that.... Write PHP scripts or arse about on PH.... Tough call

Is you dissing my PHP skills?

[The Griffalo]

mystomachehurts said:

The Griffalo said:

Hmm, let me think about that.... Write PHP scripts or arse about on PH.... Tough call

Is you dissing my PHP skills?

Nope, mine

[fastfreddy]

sp60 said:

just out of curiosity, why does it have to be rack mountable?
You could look into bladeservers as these are very small, but i think you need the main cage to slot them in to, which will probably make them over budget.
You can just buy trays to bolt into racks that allow anything to be put on them...such as a standard tower pc.

There's another piece of kit these PCs control and that's a 4u rackmounted device. It makes for a nice neat installation if we have a 6-8u rack with everything together with good cable management for reliability. Agreed that a blade server could be better in some situations, but not for this one IMO. Could put towers on shelves in the rack, but I don;t personally feel it would look as professional and tidy as an all rackmounted solution. The only drawback I have found so far with the first system we built here is the additional fan noise as has been discussed.

If I've got Kell as a fall-back solution then I'm not so worried.

[cuneus]

The main reason I left my last job (4 yrs ago) was because management specified Dell

We did buy some HP(Compaq's) when the Dell's failed.

Even with Uber level support the tech's were hopeless

Every brand has problems but at least HP can spell engineering

[LivinLaVidaLotus]

theboss said:

fastfreddy said:

glass fronted rack

I wouldn't if I were you - they pull cool air in from the front

Our racks with solid front doors cool better than the mesh ones, as air comes up through the floor at a decent pressure, it makes its way up the front of the rack and is sucked in through the front of the servers. Unless you're running a properly separated (as in physical) hot/cold isle setup I wouldn't worry about having solid doors.

All 1U servers are noisy. IBM build very nice servers, not too bad prices these days. Personally, we use all Tyan servers these days.

[minimac]

We use HP Proliants, very very good. We have some insane disk usage, so we RAID some 15000rpm drives.

We also build high spec PC's, then just pop them in rackmountable server cases, bout 70quid each.

[fastfreddy]

Thanks for all the help!

I think I've come to a decision based on the posts here and doing a bit of research. I'm thinking DL380 G5's as they should be quieter that the 360 and seems to have a very good reputation for reliability. Not a problem that they are 2U as I can still get all the kit I need for a system in an 8U rack.

Kell do a 12u enclosure which is apparently almost silent once the doors are closed. I'm going to see one next week and this could be a good option if I have to install a system in an office environment.

Still not sure whether to run Server 2003 or XP Pro. The main app will run on either apparently.

[theboss]

fastfreddy said:

The main app will run on either apparently.

I think you just answered your own Q... you ought to be using a fully certified and supported hardware and software platform... 2003 is supported by HP on DL kit (XP isn't) and, vice versa, the DL kit is on Microsoft's hardware compatability list for 2003 (but not XP).

[fastfreddy]

theboss said:

fastfreddy said:

The main app will run on either apparently.

I think you just answered your own Q... you ought to be using a fully certified and supported hardware and software platform... 2003 is supported by HP on DL kit (XP isn't) and, vice versa, the DL kit is on Microsoft's hardware compatability list for 2003 (but not XP).

Sound advice

[Morningside]

I can get DL360s for pennies from work. Now apart from the noise (Try running 10). Would or could they make fast home computers? 2xdualcore, tonnes of ram add a quality graphics card and ....

or are they best suited to server stuff.

[JamieBeeston]

Morningside said:

I can get DL360s for pennies from work. Now apart from the noise (Try running 10). Would or could they make fast home computers? 2xdualcore, tonnes of ram add a quality graphics card and ....

or are they best suited to server stuff.

They'll run a Home OS (Windows XP) without issue, but dont discount the noise so readily, they REALLY are noisey!

also the shape can be a pita!

J

[ATG]

Are you going to be able to actually test your system's performance on this new hardware before you ship it to your client?

[fastfreddy]

ATG said:

Are you going to be able to actually test your system's performance on this new hardware before you ship it to your client?

It will all be built & configured here then repacked and sent to Oz

[randlemarcus]

Morningside said:

I can get DL360s for pennies from work. Now apart from the noise (Try running 10). Would or could they make fast home computers? 2xdualcore, tonnes of ram add a quality graphics card and ....

or are they best suited to server stuff.

When you say pennies, how many can you get hold of?

And Jamie, how much is a couple of U of co-lo these days? Remote hands only support for resets, not ginormous bandwidth?

[ATG]

fastfreddy said:

ATG said:

Are you going to be able to actually test your system's performance on this new hardware before you ship it to your client?

It will all be built & configured here then repacked and sent to Oz

Cool. Then that will give you a chance to troubleshoot it, test performance and, if necessary, change the spec?

I mention all this because I've been through a painful process that may have some similarities to the one you're embarking upon.

In my case I was shipping a software system to a company and they were responsible for providing and maintaining the hardware. I gave them a hardware requirement specification, they were supposed to respond with a hardware proposal, but they just went to the shops and bought some DL360s and some very expensive SCSI discs. And the performance was a pile of shit. "Oh but we thought that ...", "Oh, but we assumed that ..." blah, blah, blah. They always used DL360s, so they just bought some more without thinking about the detail of the spec I'd given them. SImilarly they configured RAID5 on the SCSI controllers coz that's what they always did, etc, etc. They failed to notice that one of the third party applications they needed to run on this kit required XP, but as has been said, this is not support by HP on that platform so one of the boxes ended up running Win2k. They failed to spec the batteries required by the SCSI RAID controller to enable cached writes (I eventually debuuged that for them courtesy of asking some questions here on Pistonheads ... I think Mr Beeston helped out, for which i am eternally grateful). Without the cache, the write performance is dreadful. Even so, our software regularly fills the write cache so we still get shit write performance ... as in substantially slower than on the workstations we used for development. They also failed to notice that although iLO cards are fitted as standard on the hardware, you have to get an additional license in order to use it. The final thing they ed up was getting lots of mediocre CPUs in the box rather than a couple of blindingly fast ones ... their database software can't take advantage of multiple processors ... so what is the ing point of having lots of slow ones?

I am quite sure there is nothing wrong with DL360s, and if we were using something other than RAID5 on their default SCSI controller, we'd probably be able to get some blinding performance out of the many, many thousand of pounds worth of discs that was fitted.

The lesson I have taken from this is that when people rely on their own rules of thumb estimates when it comes to system performance for a specific application, they are taking a big risk. (I.e. if someone blithely says "you need SAS, not SATA" when all you have said is "disc intensive", I start hearing alarms bells. What does "disc intensive" mean? Shifting between disc systems, even when at first glance it seems "obvious" that you're leaping up the performance curve, can actually cripple you. I'd similarly have absolutely no faith at all that shifting between O/S versions would be benign. SImilarly, throwing lots of processors at a solution is utterly pointless if the system can't take advantage of multiple processors ... blindingly obvious point, I know, but it has been a real learning experience watching otherwise intelligent people make such a complete hash of some pretty basic decisions.

There are a sufficient number of factors at play when speccing hardware that I would not have much faith in performance estimates. I'd want to see real world performance figures, i.e. running your software on your proposed hardware, with a detailed breakdown of the various loads on the system so I could see the real world cost/benefit of the various features of the hardware.

Edited by ATG on Monday 20th August 10:52

[JamieBeeston]

randlemarcus said:

Morningside said:

I can get DL360s for pennies from work. Now apart from the noise (Try running 10). Would or could they make fast home computers? 2xdualcore, tonnes of ram add a quality graphics card and ....

or are they best suited to server stuff.

When you say pennies, how many can you get hold of?

And Jamie, how much is a couple of U of co-lo these days? Remote hands only support for resets, not ginormous bandwidth?

Colo costs are mostly down to power these days..

Alas in london you're looking over £130 a month per amp (typical decent spec server these days pulls 0.6-1.6a)

You can get it much cheaper out of london if cost is your prime concern..

If you want a quote, let me know and I'll have sales drop you an email, but it's actually cheaper to rent these days.

J

[theboss]

ATG said:

The lesson I have taken from this is that when people rely on their own rules of thumb estimates when it comes to system performance for a specific application, they are taking a big risk. (I.e. if someone blithely says "you need SAS, not SATA" when all you have said is "disc intensive", I start hearing alarms bells. What does "disc intensive" mean? Shifting between disc systems, even when at first glance it seems "obvious" that you're leaping up the performance curve, can actually cripple you. I'd similarly have absolutely no faith at all that shifting between O/S versions would be benign. SImilarly, throwing lots of processors at a solution is utterly pointless if the system can't take advantage of multiple processors ... blindingly obvious point, I know, but it has been a real learning experience watching otherwise intelligent people make such a complete hash of some pretty basic decisions.

Whilst I agree with your post in principle, with its emphasis on the importance of real-world testing versus assumptions and rule-of-thumb, I'm rather curious as to why a blithe recommendation of SAS disks over SATA, in response to an obvious requirement for disk performance, causes you to feel alarmed (seeing as I made that precise recommendation to the OP, somewhat blithely, myself). In any like-for-like array configuration, SAS disks will outperform SATA in every possible respect, so how is this poor or ignorant advice?

[ATG]

theboss said:

ATG said:

The lesson I have taken from this is that when people rely on their own rules of thumb estimates when it comes to system performance for a specific application, they are taking a big risk. (I.e. if someone blithely says "you need SAS, not SATA" when all you have said is "disc intensive", I start hearing alarms bells. What does "disc intensive" mean? Shifting between disc systems, even when at first glance it seems "obvious" that you're leaping up the performance curve, can actually cripple you. I'd similarly have absolutely no faith at all that shifting between O/S versions would be benign. SImilarly, throwing lots of processors at a solution is utterly pointless if the system can't take advantage of multiple processors ... blindingly obvious point, I know, but it has been a real learning experience watching otherwise intelligent people make such a complete hash of some pretty basic decisions.

Whilst I agree with your post in principle, with its emphasis on the importance of real-world testing versus assumptions and rule-of-thumb, I'm rather curious as to why a blithe recommendation of SAS disks over SATA, in response to an obvious requirement for disk performance, causes you to feel alarmed (seeing as I made that precise recommendation to the OP, somewhat blithely, myself). In any like-for-like array configuration, SAS disks will outperform SATA in every possible respect, so how is this poor or ignorant advice?

Didn't say it was poor or ignorant, just that without knowing what the OP means precisely by "disc performance" it is hard to make any recommendations at all. SATA's bandwidth is huge, so if the "high performace" is required for sustained reads and writes the discs become the bottleneck, not the controller and SAS may not offer any real benefit. As already said, I've seen the wonderful example of a server's SCSI RAID controller underperforming a workstation's IDE disc system.

[theboss]

ATG said:

Didn't say it was poor or ignorant, just that without knowing what the OP means precisely by "disc performance" it is hard to make any recommendations at all.

Fair enough, but not making any recommendations wouldn't have really helped the OP, and I still believe that a beefy SAS controller with BBWC and multiple high-rpm spindles is a sensible starting point, irrespective of the actual read/write characteristics in question.