[sims] Sims Server Performance Benchmarks

[matt40k]

The thing is, it's .NET, so it does the memory management it's compilated JIT - just-in-time. So unlike C programs which load the entire program at once, .Net and Java will load as it needs it, so the footprint is pretty small, but it will grow, so although you've opened and closed windows, it's been loading more and more modules as it goes and it's not likely to unload them once they are load. So once you've opened SIMS, go into your register, then into system manager then into reports and so on, closing each time will return a bit of memory but not as much as closing SIMS then re-opening for each area. I'm not saying this is a good idea, closing SIMS for each area, it's a bad idea. You should notice that once you've gone into edit marks it's faster to return to it because it's already loaded the .net library and built the cache. Each time you close SIMS you have to do authentication request, a pull of the school diary, check you messages, load all the homepage guff, build the cache then load whatever module your using.

 

Closing SIMS at the end of the day makes sense, closing SIMS when you go to lunch makes sense, closing SIMS when you make a big change makes sense. Closing SIMS every hour or less is like painting your computer red because it'll go faster.

[vikpaw]

I only advise it where it works, and it's usually admin staff who are doing big jobs, or spending all day entering new student applications. So it is the right thing to do.

Most staff change class rooms every 90 mins, so it's automatic.

I see what you mean. Since that post, the RAM usage has halved, but it is still 50% more than it was while idling all morning. I guess it's happy with sporadic use.

 

The bulk of server side performance was introduced with the last few updates, so they have definitely changed something, and there is probably a lot more tweaking that could be done, not of the server / hardware but the server side software. A lot of issues were attributed to the homepage widgets which can still take their toll.

[matt40k]

I should really dig out the reply email I got when I asked what the new homepage widgets performance impact were when I tested it in beta all those years ago.

[localzuk]

The bulk of server side performance was introduced with the last few updates, so they have definitely changed something, and there is probably a lot more tweaking that could be done, not of the server / hardware but the server side software. A lot of issues were attributed to the homepage widgets which can still take their toll.

 

Indeed. As @jinnantonnixx mentions, it appears they've been doing more database work lately. Which would account for quite a bit. Also, .Net is becoming more mature, and performance improvements have been made by Microsoft. So, its quite possible that some of the speed increases are down to Microsoft too.

[psydii]

As vendors have started to offer hosted solutions they are experiencing the pain of their design choices first hand, which is forcing them to make improvements that otherwise they would let slide. Building reports seems to be slow in most MIS's. A good one will let you get on while it builds and then emails it to you when done.

 

Also dotnet has been very hard to debug and tune for performance until quite recently (or so i read the inference to be from "Defrag Tools") so any performance issues may just be that banks have got all the best developers and those of us who think £15K is a lot of money for an 1500 seat organisations main LoB application get the left overs.

 

That said we have got a locally hosted web based mis with 7 years of data in it and it is lightening fast (though reports take a while to build).

[vikpaw]

Indeed. As @jinnantonnixx mentions, it appears they've been doing more database work lately. Which would account for quite a bit. Also, .Net is becoming more mature, and performance improvements have been made by Microsoft. So, its quite possible that some of the speed increases are down to Microsoft too.

Actually i meant performance impact (negative) - but yes, along with that, they have also improved as you say to mitigate that effect.

 

I've always thought companies should use their own products. e.g. if you're selling a VLE, use the built in chat room to do a sales meeting, not Skype or WebEx.

If you're selling a mailing solution send via that, not Exchange. Obviously doesn't work for all products, but use them where you can, cos it shows the weakness to you if it exists, and shows off the solution to the customer which is great.

[zag]

 

Much better CPU performance on Autumn 2 release.

 

Well done Capita.

 

The program is still slow loading and solus is very unresponsive, but its an improvement from the last release.

[vikpaw]

We had a few problems recently with our power supply to the server room.

This had a couple of very unusual side effects, at least they were very unexpected.

 

After a power loss and some issues with the backup generator, we found a few places had no or reduced power. We had to take a lot of systems off the UPS which no longer would accept an under voltage supply - due to a different issue. The result was that a lot of the systems started under performing.

 

I'd never experienced such issues, but the ping response on a lot of servers went haywire and became very unpredictable and a lot of network traffic seemed to slow.

 

Specifically, some SIMS reports would just hang and not even crash out which was when i got involved. I could run the reports with a query that i knew would only return very few results and they would be fine, so it started timing out sooner as well as responding slower: load times were massively sluggish, for the first open of app as well as individual modules.

 

Another thing we noticed was that on our blade chassis the VMs on a couple of the blades had better performance than others, and moving the SIMS VM to the "better" blade, improved things, but not totally. We later found out that some auto safety feature in the blade had detected low power and throttled some of the blades. Shame it wasn't configured to tell us, and we had no control over it!

 

Anyhoo, all is back to near normal now, yet we still have a slight under voltage and it does still have a slight effect. The flow isn't smooth here and the UPS definitely makes a difference, we got a new one in the end. However, generally the other items that weren't on the UPS and the system as a whole were largely affected by the power.

 

Power quality was a big factor, obvious when you think about it, but not something we'd ever have considered had we not had a failure, and with a working UPS, we might not even have noticed for a long time.

 

So when doing these tests and benchmarks, it's worth thinking outside the box. I've never really experienced the slowness others talk about with SIMS, and wouldn't have thought anything of it, until it suddenly affected me. I probably would have quite happily blamed Capita for shoddy programming, which in some cases might be true, but you don't often think the network is to blame when things are working ok. I can't compete with the likes of @bossman 's network and speeds, but we're pretty good now, even with a poor supply.

 

Hope that helps the odd person or two.

[synaesthesia]

We're talking with Capita a lot currently due to performance issues that have gotten far worse since the Autumn upgrade. When someone (data manager for example) is running a large report it generally makes it impossible for other users to log in, or if they can it's slow. Slows other users down to a crawl if not a halt.

Reindexed, shrunk the log files but it's still a bit silly performance-wise. Considering the sheer grunt of our server (it's not even virtualised, we're effectively wasting it) and the capableness of the network I'm a little more than concerned about Capita's statement of the inevitability of it going "cloud based". I can only assume any plans to do so (are those plans still live? @PhilNeal , I seem to recall one of your guys blogging about it a couple of year ago?) would involve a lot of work internally more than just shifting it into a huge server farm as-is.

[vikpaw]

@jinnantonnixx runs or did hundreds of setups on the same SQL server. It's possible. I'm sure they've had performance issues a big lately too. Not sure if it plagues them all or just individual DBs.

[jinnantonnixx]

We're talking with Capita a lot currently due to performance issues that have gotten far worse since the Autumn upgrade. When someone (data manager for example) is running a large report it generally makes it impossible for other users to log in, or if they can it's slow. Slows other users down to a crawl if not a halt.

Reindexed, shrunk the log files but it's still a bit silly performance-wise. Considering the sheer grunt of our server (it's not even virtualised, we're effectively wasting it) and the capableness of the network I'm a little more than concerned about Capita's statement of the inevitability of it going "cloud based". I can only assume any plans to do so (are those plans still live? @PhilNeal , I seem to recall one of your guys blogging about it a couple of year ago?) would involve a lot of work internally more than just shifting it into a huge server farm as-is.

 

 

We had dire performance, and the clients regularly timed-out. The report speeds are now back to normal after setting up a maintenance plan to rebuild indexes. The trouble with the SIMS reindex patch is that it doesn't update the statistics. Rebuilding the indexes generates new statistics so the query optimiser can produce better execution plans. If Capita added a further index for att_mark_normalised things would improve further.

I've detailed this in another post about slow SIMS, link below.

http://www.edugeek.net/forums/mis-systems/128123-sims-gradually-getting-slower-slower-2.html#post1104947

Edited January 17, 2014 by jinnantonnixx

[matt40k]

@synaesthesia it'll be alot easier in a cloud environment, you not worrying about the 99 ways of the user has configured SQL, or if the hardware is the best it can be. Also you can actually run proper performance monitoring on the environment so you'll pick up some weird scenario that they've never seen before. This is assume you get a performance SLA of course!

[synaesthesia]

That's true enough.

I think it said a lot when we had a phone call from Capita to apologise for the delay in getting back to us on the open case, as they're overrun at the moment!