Page 1 of 2 12 LastLast
Results 1 to 10 of 14
Like Tree2Likes

Thread: XenDesktop/XenServer Revit poor performance?

  1. #1
    Junior Member
    Join Date
    October 8, 2014
    Posts
    7
    Current Local Time
    04:50 AM

    XenDesktop/XenServer Revit poor performance?

    Hello

    We're running a PoC for Revit in XenDesktop with nvidia grid cards.
    Dell R720, 8 300gb SSD raid 10
    Internal SSD drives

    The virtual pc specs are
    Windows 7 64-bit
    CPU: Intel Xeon E5-2690 @ 3 ghz 4 virtual cores (tried with 8, same issues as below)
    16 gb RAM
    Nvidia Grid2k K260Q 2gb ram

    Revit 2014 Update 3

    Most work our customer do in revit works fine in the VDI environment but some things are really really choppy.
    Most notable is if i try to add a Pipe, copy an object or messaure distance, everything start lagging if they do it in a big project.

    If i compare 3d view to their normal workstations its about the same.

    I ran the benchmark tool availible here and the results seem a bit on the low side i think.

    4 Cores
    Model creation and view export benchmark
    4,0 75
    14,0 57031
    39,0 63281
    72,0 28125
    33,0 67188
    19,0 40625
    14,0 60156
    36,0 17188
    47,0 75781
    278,0 Total

    Render benchmark
    322,0 6094

    GPU benchmark*
    16,0 96875
    9,0 859375
    12,0 42969
    10,0 27344

    45,0 4375
    39,0 0625
    42,0 14844
    47,0 26563

    8 Cores
    6,0 607422
    13,0 18164
    37,0 33398
    70,0 125
    32,0 5625
    18,0 26758
    14,0 75391
    33,0 26953
    48,0 57813
    271,0 Total

    Render benchmark
    184,0 9219

    GPU benchmark*
    17,0 39844
    9,0 933594
    13,0 28516
    9,0 152344

    48,0 78906
    42,0 11719
    43,0 94531
    37,0 22266


    Any ideas how to speed things up or particullary why the Pipe/copy/meassure thing seems to be such a performance consumer
    Last edited by pejter; October 17th, 2014 at 08:10 AM.

  2. #2
    Forum Co-Founder iru69's Avatar
    Join Date
    December 7, 2010
    Location
    United States
    Posts
    1,436
    Current Local Time
    09:50 PM
    Hi Pejter,

    Sorry, I can't answer your question, but just from what I've seen of similar setups, it actually seems like you're getting pretty good performance considering the poor performance these various virtual app setups generally display. Maybe someone will chime in with some tips though. Good luck!

  3. #3
    Member irneb's Avatar
    Join Date
    April 25, 2013
    Location
    Jo'burg SA
    Posts
    308
    Current Local Time
    06:50 AM
    It seems that your accelerated graphics is performing very close to an entry level discreet GPU (e.g. only around 20% slower than my GTX 560Ti), but around twice as slow as something like a Quadro K600. I'd be very surprised if you'd be able to get much more performance from that grid card than you're getting right now. This type of card is still in its infancy in comparison to "normal" GPUs, not to mention the VM manager itself is also still only a year or two in the testing - so more optimizations might be in the pipeline for the future.

    As for the edits being slow ... not sure about that too much. I'd expect a slight decrease in speed due to the VM's virtualization. From my previous tests (though only through VirtualBox) the same benchmark on bare metal as compared to through a VM (on the same machine) meant around 5% increase in time taken (this I tested a few years ago on an i7-2600). And your test seems a bit worse in this case - I'd like to see a bare metal benchmark on that same CPU to see exactly what your Cytrix VM is wasting. Especially since your CPU is a higher Hz and a bit "newer" than my current X5650 - which runs at about 80% of the time taken in yours (when running bare metal). Adding multiple cores wouldn't help much on the editing side, but would definitely help with rendering, as well as some of the non-accelerated view changes.

    BTW, is the 16GB RAM the total in the server or how much you've set for the single VM? Sorry not sure if Citrix sets RAM as static shared like VirtualBox or dynamically adjusting like VMWare.

  4. #4
    Moderator
    "RFO 2012 Spelling Bee Champion"
    Alex Cunningham's Avatar
    Join Date
    December 8, 2010
    Location
    Northern California
    Posts
    3,392
    Current Local Time
    09:50 PM
    I am having a issue with my IT wanting to configure the RAID like it needs to have 2x data backup. IHMO I don't think these virtualizing computers need to have data redundancy. I would vote for a RAID0 over the RAID10. IMO if the server crashes they can restore it for a image in a few min. There is no reason for hot swapping a drive in the event of a crash. These servers are basically workstations and nothing on them is critical to retain. But that just my opinion. Also check out https://www.citrix.com/content/dam/c...essmode=direct

  5. #5
    Member irneb's Avatar
    Join Date
    April 25, 2013
    Location
    Jo'burg SA
    Posts
    308
    Current Local Time
    06:50 AM
    Quote Originally Posted by Alex Cunningham View Post
    I would vote for a RAID0 over the RAID10.
    True. It makes no sense having your programs and temporary data folders on an aggressively redundant drive. Ask your IT guys if they think it a good idea to install RAID 1+ as standard for normal workstations also.

    Anyhow (off topic), RAID ain't all that great. It usually takes ages to regenerate "lost" or corrupt drives. And more often than not such regeneration fails anyway (i.e. backups are your best bet not RAID). Not to mention, they don't even guard against software-caused corruption (which is much more prevalent than hardware failures), i.e. if a file is saved with an error inside it the RAID would simply duplicate that error several times over.

    I've tried several RAIDS in the past, and thus far the only ones which even have an iota of usefulness is the software raids like the RAID-Z from ZFS - originally made for Solaris Unix but now available in BSD as well as Linux. None of the hardware stuff help at all, and strangely their performance is worse than the ZFS one.
    Last edited by irneb; October 14th, 2014 at 09:17 AM.

  6. #6
    Junior Member
    Join Date
    October 8, 2014
    Posts
    7
    Current Local Time
    04:50 AM
    Quote Originally Posted by irneb View Post
    It seems that your accelerated graphics is performing very close to an entry level discreet GPU (e.g. only around 20% slower than my GTX 560Ti), but around twice as slow as something like a Quadro K600. I'd be very surprised if you'd be able to get much more performance from that grid card than you're getting right now. This type of card is still in its infancy in comparison to "normal" GPUs, not to mention the VM manager itself is also still only a year or two in the testing - so more optimizations might be in the pipeline for the future.

    As for the edits being slow ... not sure about that too much. I'd expect a slight decrease in speed due to the VM's virtualization. From my previous tests (though only through VirtualBox) the same benchmark on bare metal as compared to through a VM (on the same machine) meant around 5% increase in time taken (this I tested a few years ago on an i7-2600). And your test seems a bit worse in this case - I'd like to see a bare metal benchmark on that same CPU to see exactly what your Cytrix VM is wasting. Especially since your CPU is a higher Hz and a bit "newer" than my current X5650 - which runs at about 80% of the time taken in yours (when running bare metal). Adding multiple cores wouldn't help much on the editing side, but would definitely help with rendering, as well as some of the non-accelerated view changes.

    BTW, is the 16GB RAM the total in the server or how much you've set for the single VM? Sorry not sure if Citrix sets RAM as static shared like VirtualBox or dynamically adjusting like VMWare.
    16 GB for all virtual machines. Ye i would accept it if it was slightly worse. But its so slow that they can't really use it at all. But its just these copying, or creating certain objects. One thing i noticed with my novice revit experience was that i could choose different types of pipes. The default pipe which could be created in a straight line was OK but their custom or template pipes with calculations for degrees that when it all went downhill.

    Like i wrote before, we have found 3 things that are too slow.Copy certain objects, building pipe systems and measuring distance in certain drawings. Other things work good or even better then on their workstations.

  7. #7
    Member irneb's Avatar
    Join Date
    April 25, 2013
    Location
    Jo'burg SA
    Posts
    308
    Current Local Time
    06:50 AM
    Quote Originally Posted by pejter View Post
    16 GB for all virtual machines.
    If I understand you correctly, the server has some great amount of RAM and each VM is set to use only 16GB out of that. If so then it's very much the same way VirtualBox works. VMWare OTOH shares the entire RAM between all VMs, so you see something like 192GB in some cases - even though each client is only really using a small fraction of that - i.e. if you load more stuff you get given more RAM. This is why I'm asking, because depending on the VM hypervisor the stated RAM is either that assigned to you or the total shared by all - if the later then your RAM is just about good enough for small to medium projects with only 2 concurrent client OS's (I'd be very scared of dropping below 8GB per "workstation").

    As for the CPU, that's supposed to be an 8 core hyper threaded (i.e. looks like 16 cores total). The trouble with a VM is that switching between these cores for each running client OS causes some performance degrade, though from my experience it shouldn't be a drastic drop (around 5 to 10% slower). So some issue might be from this, but at a guess you shouldn't notice much influence. And you're correct about not noticing any difference by adding more cores - for that matter you'd probably see very little diff if you only used 2 cores (you might find a bit of an issue if only one core though).

    BTW, I'm not too familiar with Citrix, so not sure if same could be said about it. Got much more experience with VMWare / VirtualBox. In VMWare you can opt for 2 ways of using the passthrough graphics (i.e. when not using the software 3d): vSGA and vDGA. The S shares the graphics card amung numerous users, the D shares each GPU to one user only.

    On the Grid K2 you have 2x "high" end GPUs, so if you set the vDGA method you can only have 2 concurrent users with hardware accelerated graphics. Else with vSGA you can have several, but the performance would suffer. The Grid K1 is a bit different, it gives 4 discreet GPUs but each are lower speced than the ones on the K2. The easiest way for a user to see which is used on the server - with vDGA the card's own driver needs to be installed in the client OS, with vSGA the VMWare graphics driver is installed instead.

    Quote Originally Posted by pejter View Post
    Like i wrote before, we have found 3 things that are too slow.Copy certain objects, building pipe systems and measuring distance in certain drawings. Other things work good or even better then on their workstations.
    Strange that you've only got horrors happening in some circumstances - have you tried using Process Monitor to see if it's not something wrong with the setup? E.g. some missing file or registry key?
    Last edited by irneb; October 14th, 2014 at 03:11 PM.

  8. #8
    Junior Member
    Join Date
    October 8, 2014
    Posts
    7
    Current Local Time
    04:50 AM
    Yup thats right 256gb of ram in the host and 16gb to all the virtual windows 7s.
    For cpu we got two 10 core hyperthreaded xeon processors. So we have 40 virtual cores to play with. And for the vdi i've configured it so that it only uses cores from the same socket.
    Gpu we have virtual Gpus no passthrough. Two grid 2k cards on the highest gpu setting k260Q.
    If i look at the performance graphs in xencenter or resource monitor theirs no issues, at least not what i see.

    The host is on the same network as their workstations.

    Yup i've tried using procmon, the only weird thing i notice is that revit seems to look/create files into Windows\CSC which is the folder for offline files. The function is disabled tho, and same issue with it enabled.

    I've also noticed that adding anything from the "systems" tab in revit ducts, pipes, conduits wc objects, is slow. But if i add a wall for example from the architecture tab that works fine.

    if i understand correctly, every time i add an object, revit takes all the other objects into consideration when calculating stuff?
    I've tried the cropping function, selecting only a tiny part of the structure, same issue there.
    If i open a new project, everything works really smooth.

  9. #9
    Member irneb's Avatar
    Join Date
    April 25, 2013
    Location
    Jo'burg SA
    Posts
    308
    Current Local Time
    06:50 AM
    Quote Originally Posted by pejter View Post
    I've tried the cropping function, selecting only a tiny part of the structure, same issue there.
    If i open a new project, everything works really smooth.
    That tells me there's some issue with the project itself. Most usually it's due to ignored warnings. Go to the Manage tab and click the Warnings button ... how many of them are listed?

    It's recommended (actually that word's a bit "soft" for this case: "you should attempt to ALWAYS" is a bit closer to the importance of this) fix any warnings and periodically review any of those left, trying to fix them. The trouble with warnings is that Revit constantly goes through them on each and every edit - sometimes slowing a project down to a crawl.

    But this makes no sense, as I understood it you had none of these issues when editing on a bare-metal workstation? If this is the case then I'm not sure what else could be going on. If the same trouble is seen when not through a virtual environment (i.e. on a normal workstation), then it's most probably a "Keyboard-Seat-Interface-Error"
    Last edited by irneb; October 15th, 2014 at 09:28 AM.
    iru69 and Alex Cunningham like this.

  10. #10
    Junior Member
    Join Date
    October 8, 2014
    Posts
    7
    Current Local Time
    04:50 AM
    This is how the warnings look if i open one of their projects that is slow.
    Theres a total of 1196 warnings under those various categorys.

    Many intresting, like It is not possible to calculate the flow since the flow direction mismatch.
    There are identical instances in the same place. this will result in double counting in schedules.

    Should it be 0 warnings :O?

    This doesnt either explain why it works on their pcs, but if they fix their issues maybe it goes smoother.
    Attached Thumbnails Attached Thumbnails XenDesktop/XenServer Revit poor performance?-warnings.jpg  
    Last edited by pejter; October 15th, 2014 at 12:30 PM.

Page 1 of 2 12 LastLast

Similar Threads

  1. Revit Performance and Microsoft Outlook
    By buildbod in forum Architecture and General Revit Questions
    Replies: 8
    Last Post: May 20th, 2013, 08:25 AM
  2. Replies: 0
    Last Post: November 20th, 2012, 01:16 AM
  3. Replies: 0
    Last Post: February 21st, 2012, 03:15 PM
  4. Revit 2012 & 3D performance...
    By muttlieb in forum Architecture and General Revit Questions
    Replies: 7
    Last Post: October 12th, 2011, 04:06 PM
  5. Revit Performance Drop Off
    By Alex Cunningham in forum Hardware and Infrastructure
    Replies: 14
    Last Post: April 16th, 2011, 03:05 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •