An infrastructure startup has had to shift off of cloud servers to bare metal gear after it ran into performance issues, showing that virtualization is not always the optimal tech for cloudy apps. Firebase, a backend-as-a-service company that provides a technology which lets developers sync JSON data in real time across …
needs real computers.
If you are running LAMP, and each user has web applications + data base, which is better performance?
1 copy of SQL & Apache and Linux with N user accounts
N x VMs, each with a copy of Linux, Apache and SQL?
*nix OS family was multiuser from the beginning
Only if you have users installing native applications is the overhead of VM needed.
For 99.99% of ordinary hosting the "VM" is a waste of money and performance.
VMs and Cloudy Computing has it's niche. But it's stupid for most things.
Re: Someone somewhere
It really depends on what those accounts are doing. In scenario 1 you can get the best overall performance but that is not the only consideration -- try co-ordinating downtime for all the customers sharing that system for things like OS updates, major database changes etc. What if customer X needs some patch installed that breaks customer Y's systems? What if customer C is doing a bunch of bad sql and causing massive locking impacting the DB's ability to serve other customers? In the case of oracle (locking related too just had a flashback) what if customer G runs a bunch of stuff that causes latch contention? (basically drives cpu to 100% and throughput to zero).
For 99.9% of ordinary hosting VMs provide more than adequate performance, because most(I'd wager at least 95%) systems sit at less than 10% utilization at any given point. So you can take a rack of servers running VM software, or you can take 50 racks of other servers providing the same service - which is more cost effective? An alternative approach is to use micro servers, but those in themselves are even more limited (memory/cpu configurations). If your app(s) work well with those configurations then by all means have at it.
At the end of the day it comes down to determining the best for the application(s) at hand. Some are very well suited for bare metal, most others it would be a waste of energy, and hardware capacity.
But as for cloud - yes it is pretty bad. People should not confuse public clouds with what customers can achieve on their own private infrastructure (whether you call it cloud, or I refer to it as utility computing). The two are further apart than virtual vs physical.
cloud vs virtualization
you can't really put a blanket blame on virtualization when a cloud service provider fails.
It's not rocket science to run a virtualization infrastructure with good performance and low latency. Especially if you are controlling everything end to end. Having full insight into what everything is doing is key (this is often(almost always) not possible with cloud service providers).
However if you can leverage the underlying hardware and perhaps can run multi tenant from within a single app then you may be better off with bare metal. So many people seem to think that bare metal is a terrible thing to do these days, but it can make sense if your app can leverage it right. Otherwise put a hypervisor on there and run multiple VMs to better utilize the hardware.
I was talking with another startup recently that is a service provider of sorts, they too started out in the cloud (various clouds) but were forced to move out quickly because of latency and lack of consistent performance issues (cost too but that wasn't the primary driver I gather).
The solution is often only as strong as the weakest link, if your lucky and have a robust software stack then the next weakest link is probably the infrastructure, and the most commonly used cloud infrastructure is not built to be very reliable.
::eyeballs the running Tandem NonStop VLX in the machineroom/museum/mausoleum/morgue::
::takes dawgs out or a pee, muttering about kids these days::
You have to reboot your own server now?
Re: You have to reboot your own server now?
Yeah, wish physical servers booted as fast as they did in the more recent Mission impossible Film
**The world is saved**
- 20 Freescale staff on vanished Malaysia Airlines flight MH370
- Fee fie Firefox: Mozilla's lawyers probe Dell over browser install charge
- Neil Young touts MP3 player that's no Piece of Crap
- Review Distro diaspora: Four flavours of Ubuntu unpacked
- Sysadmins and devs: Do these job descriptions make any sense?