The problem with these types of shared resources are what they are, that is they are shared. So many times I have seen customers who have invested heavily on great equipment like NetApp FAS units or Dell EqualLogic believing all of their storage related issues will be resolved. In order to guarantee a fast desktop you need to be able to be in control of your back end resources. So there you have it, for SBC\XenApp Desktops you can each server will cope with 200 desktops and for VDI that number comes down to 40. The choice to select what desktop to run VDI vs SBC will be covered in a separate paper. Making the Hardware Cost Per VDI Desktop = £90 per user (quite a difference) Let compare this with VDI\XenDesktop\VMWare View the numbers will be different because we are publishing many more operating System but this time we will be running Window 7\8 each with 2GB RAM, 1 vCPU and 25 IOPS and as we can see from the chart above you theoretically get 58 desktops running off one server but then apply our 1 third contingency we are looking realistically to achieve around 40 Desktops per physical server Making the Hardware Cost Per XenApp Desktop = £32.50 per user
User logons are critical metric also to consider and these servers are capable of handling 14 simultaneous logons at 100 IOPS with a 1440 IOPS capacity On this VM you you will comfortably get 35 users sessions running concurrently and on each physical server of the specification above will run up of eight of these VM’s on the same server giving you a total of 280 sessions per HP Servers but for peace of mind size it up with 200 user session in mind, knowing there is spare capacity with will be important later. So If you’re running a XenApp\SBC\RDSH solution we have found that the optimum hardware configuration for each VM is 4 vCPU’s, 12GB RAM, 100GB HDD with a single NIC that is part of multi-NIC bonded network of four NICs. What this all comes down to is how many desktops\sessions can I run per physical servers?Īs you can see from the chart above Disk IOPS are our biggest bottleneck, not CPU and Memory. In order to make sense of this we need to translate this down to physical hardware and for the sake of simplicity here I am going to take an average run of the mill mid-range HP Server, say a D元80 Gen 9 with 2 x ten core 2Ghz processors, 128GB RAM with an additional 4 x 1GB NIC along with 8 x 15K 300Gb SAS drives and 2GB Cache RAID Controllers at the cost of around £6,500 What does this give me?
No more than 4 users per CPU Core with HT considered.No more than 14 users per CPU Core with HT considered.2GB RAM per session server and 200MB for each session.If you can allocate theses resources to your Citrix sessions I will guarantee it will pretty fast. Here are the magic numbers that you need to remember: * Exceptions considered here are if you using products like Atlantis ILIO or Nimble where disk IOPS are measured in the hundreds of thousands. Reduce the logon time to be as fast as can be because the servers is under its greatest load during logon\logoff.Make sure the File Servers, Print Servers and user profiles are in the same subnet as the Citrix servers, especially when using profile redirection settings.Don’t bother splitting up the hard drives unless they can be spread across different dedicated LUNS, ideally you won’t be using a SAN for the Citrix Desktop (Covered later in this paper).Do not place end user sessions or desktops on equipment that is shared with other systems where you cannot ring-fence the required resources, examples being a shared SAN on VMWare*.I won’t be covering GPU or High end power users, those I will cover in a later paper. Of course it does depend on what you are doing with Citrix as the numbers will vary greatly but I am talking about the masses here. There is no black art when it comes to Citrix Capacity Planning, it is in fact very simple math. TCP Offload (under certain circumstances).There are so many things that can cause a Citrix session to slow down but the most obvious are covered here: I have a design methodology that simply states that performance can be guaranteed with good design and to me this is simple, never share and never over allocate backend hardware resources and providing you do your math right in your capacity planning and you stick to the best practices covered in this paper you can almost guarantee a fast desktop that in many cases one that will outperform a local PC. This session is all about performance considerations when using Citrix along with some best practices to prevent steady state related performance bottlenecks. It has to be one of the biggest complaints I hear about Citrix, slow logons, sluggish sessions with typing delays, poor video and sound.
0 kommentar(er)
