happy new year! I've been referring to the SSS "ma...
# citrix-vad
o
happy new year! I've been referring to the SSS "magic multiplier" rule of 5 & 10 since first reading about it in 2017 via blog posts from the mighty ex-Citrix employee Nick Rintalan1 Citrix eventually integrated the SSS method into their tech zone articles: https://community.citrix.com/tech-zone/design/design-decisions/single-server-scalability/ Now, I realize this method for calculating host capacity isn't perfect, but until Dec of 2024, i've never really had it NOT hold up in the real world, where I had a customer noting performance issues, but everything on the host side indicated their Citrix published apps setup SHOULD work with the hardware they purchased. So, new year, new questions! Do people still use the "rule of 5 and 10" for helping to plan capacity or resolve new performance issues? If not, why not, and what do you use in it's place ; outside of load testing software like LoginVSI, Lakeside
n
Been a while since I went through this routine myself, but Citrix had a session at an old Synergy event (RIP) where a group showed a way to calculate scalability to a decent degree. Citrix naturally pulled the session from their YouTube channel, but I can at least show you screenshots of their methodology.
💯 1
Hell, I have my old spreadsheet when we were planning a hardware upgrade, too, if it helps.
o
Awesome! Yeah, DM if to me if you had it. In the end, I was able to calculate it via XLS on my own yesterday, but it would be good to look at alternative methods that aren't using 3rd party software
n
This is like 3+ years old at this point, and you'll have to look at the formulas to get the calculations, but it's not terrible.
We had a mix of large and small VMs, so if you run one size, scalability is easier to calculate.
💯 1
l
Were you able to hit a 5:1 CPU ration on the XD? That seems really high relative to what I've seen. Even on the latest Ice Lake procs the best I've seen is approaching 4:1 but just a bit under. What kind of users were they? Even the call center folks I was building for would just obliterate things with hundreds of chrome tabs, and poorly designed apps just ate cpu.
o
@langsbr I agree! The rule of 5 & 10 is very optimistic. TBH, a lot of times i'd come into a customer environment, and see that they were already running around the '4' mark, and running into issues. For my most recent customer, they were running at an average of just 2.3, but were already running into issues. They were running very taxing accounting software. So, we've recommended they add more hosts to bring down the ratio BELOW the 2:1 ratio
l
@Owen Reynolds I was looking at that design decision doc a few weeks ago and thought it was crazy that they would recommend or suggest a 2 core VM for a desktop now days. Everything I've done has been 4 cores and that's been for at least the past 4 years. Heck, my old VP wanted us to just give users 8 or 12 cores and it took forever to try and explain CPUREADY effect to him (the dinner reservation thing works best!).
What kind of hardware do they have? The latest Ice Lake procs are lightyears ahead of previous generations and they are the only ones I'd even try 4:1 on. I just got my current place to order some AMD Epycs with 2x64 core procs and I want to see how these scale
n
Ironically enough, I never got a chance to see how the numbers worked out. They dragged their feet on the hardware refresh and I left right before they placed the order.
That said, we had no intention of running them that hot. The listed numbers were a 'worst case scenario' type deal.
k
The role of 5 is super basic when it comes to sizing, it never took hw refreshes into account and that makes a world of difference
o
Great discussion, all! Thanks for your inputs and have a great weekend 🙂