Terminal Server (Remote Desktop) hardware sizing
Hello, I need to size a Remote desktop server for a client. I've not done this before, ergo I am unsure what is recommended server sizing for processors, spindles, RAM, NICs etc. What are the key metrics?
So, here are the requirements:
Server 2012R2 (running as a backup AD server and DNS)
Currently 30 Users in Active Directory, 25 will be TS users
Client OS: Vista, Win7, Win8
Users run Office applications, webbrowsing, and low overhead LOB apps.
May switch to thin clients or retask older machines
2TB user data (growing)
Thank you very much for your thoughts and recommendations.
Johannes Banck
So, here are the requirements:
Server 2012R2 (running as a backup AD server and DNS)
Currently 30 Users in Active Directory, 25 will be TS users
Client OS: Vista, Win7, Win8
Users run Office applications, webbrowsing, and low overhead LOB apps.
May switch to thin clients or retask older machines
2TB user data (growing)
Thank you very much for your thoughts and recommendations.
Johannes Banck
SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
Hello:
Thank you for your thoughts on this.
I was looking at the HP ML350p Gen 8 with two of the low-end processors with each 16GB RAM (total32GB)
P420i controller board (builtin) with 1GB smartcache
2 SAS 10K 146GB mirrored drives for OS
3 SAS 10K 300GB RAID5 drives for data gives easy expandability
As configured without OS its $7K
Somehow, I am thinking it's overkill.
The company has been in business for a long time and currently have about 600GB data. The users are not high activity - how much can there be with Office apps and email?
Maybe 1 processor could be enough but with 32GB. I am almost thinking that 16GB could be enough: 5GB (x2) for User apps and 6 to 10 GB for OS. I calculated 30 Users x 200MB ea (high average).
I think I rather throw the money into Disk IO (the420i controller w. smart read/write cache).
What is your take?
Thank you for your thoughts on this.
I was looking at the HP ML350p Gen 8 with two of the low-end processors with each 16GB RAM (total32GB)
P420i controller board (builtin) with 1GB smartcache
2 SAS 10K 146GB mirrored drives for OS
3 SAS 10K 300GB RAID5 drives for data gives easy expandability
As configured without OS its $7K
Somehow, I am thinking it's overkill.
The company has been in business for a long time and currently have about 600GB data. The users are not high activity - how much can there be with Office apps and email?
Maybe 1 processor could be enough but with 32GB. I am almost thinking that 16GB could be enough: 5GB (x2) for User apps and 6 to 10 GB for OS. I calculated 30 Users x 200MB ea (high average).
I think I rather throw the money into Disk IO (the420i controller w. smart read/write cache).
What is your take?
Two CPUs is definitely overkill, people run about 20 virtual servers on a single CPU. Don't go for high clockspeed, you pay a lot for the testing and the slower marked processors will still clock at high speed for short bursts. Thread count matters since they're all waiting on disk I/O so many cores are good.
Just a couple of disks for the OS will do, or is this your fileserver too? A pair of 1TB 7.2K disks would probably outspin 3 off 10K 300GB for data just because the RAID 5 write overhead is eliminated although admittedly the FBWC improves RAID 5 speed dramatically.
Just a couple of disks for the OS will do, or is this your fileserver too? A pair of 1TB 7.2K disks would probably outspin 3 off 10K 300GB for data just because the RAID 5 write overhead is eliminated although admittedly the FBWC improves RAID 5 speed dramatically.
ASKER
Hello hecgomrec:
It is an interesting thought about going really thin (not even RAID5) on server harddrives and instead rely on backup in case of catastrophic failure of the server drives. In order to do this one must be truly confident in the DR solution.
I am a dattobackup reseller and as such I am supremely confident in the reliability of the technology - but despite that, I would still be reluctant to take safety out of the server.
Have you actually implemented such a thin solution?
It is an interesting thought about going really thin (not even RAID5) on server harddrives and instead rely on backup in case of catastrophic failure of the server drives. In order to do this one must be truly confident in the DR solution.
I am a dattobackup reseller and as such I am supremely confident in the reliability of the technology - but despite that, I would still be reluctant to take safety out of the server.
Have you actually implemented such a thin solution?
What do you mean "not even RAID 5"? RAID 1 or 10 is far more reliable and faster than RAID 5, it just doesn't give the highest capacity so was useful a while ago when disks were maxed out at 36GB. With today's huge capacity disks you should be looking at multiple disks mirrored in case one fails and another has bad sectors on it.
ASKER
Hello andyalder:
Thank you for seconding my suspicion that the Disk IO is more important than CPU on TS and file servers.
I usually go with dual RAID1 setups for reduced risk as you suggested in cases where I have a firm handle on data-growth. In cases where data growth is uncertain, RAID5 seems more elegant.
There is in my opinion no speed advantage in RAID5 arrays with less that 5 disks.
Thank you for seconding my suspicion that the Disk IO is more important than CPU on TS and file servers.
I usually go with dual RAID1 setups for reduced risk as you suggested in cases where I have a firm handle on data-growth. In cases where data growth is uncertain, RAID5 seems more elegant.
There is in my opinion no speed advantage in RAID5 arrays with less that 5 disks.
RAID 5 certainly is an elegant algorithm, but I would say it has had its day.
Data growth is just as well covered by RAID 10 as with RAID 5 although admittedly you have to add two disks at a time to the array to expand it. RAID 1 and 10 are identical on Smart Array controllers so you can start off with just 2 and add 2 more later to avoid the RAID1->RAID5 migration of adding a single disk to the array.
Data growth is just as well covered by RAID 10 as with RAID 5 although admittedly you have to add two disks at a time to the array to expand it. RAID 1 and 10 are identical on Smart Array controllers so you can start off with just 2 and add 2 more later to avoid the RAID1->RAID5 migration of adding a single disk to the array.
To JBanck.... yes I don't waste resources in a server that won't hold mission critical data!
Like users stations!.... if their data is in a well configured hardware failure server why should I bother to have such an expensive solution when I can just restore/copy their stations config.
Anyways!! this is only for a small environment... bigger the environment bigger the possibilities of failure just by high access levels to the hardware... so go on that...
Like users stations!.... if their data is in a well configured hardware failure server why should I bother to have such an expensive solution when I can just restore/copy their stations config.
Anyways!! this is only for a small environment... bigger the environment bigger the possibilities of failure just by high access levels to the hardware... so go on that...
ASKER
Thank you all, I have now a better handle on the hardware requirements for a TS server.
Regards, Johannes Banck
Regards, Johannes Banck
I would second Machienet's suggestion of a ProLiant but if you're going for a rackmount model I'd go for the DL380 rather than DL360 so you have more disk bays if needed. Alternatively a tower may be better suited especially if you want to do good old tape backups.