topic Re: HP Gen9 Server in Scalability

HP Gen9 Server

ordre — Tue, 24 Mar 2015 08:44:26 GMT

Dear All,

Do you have any chance to test or run QlikView on HP gen9 server? We have been prefering HP gen 8 servers. But now I wonder Gen9 performans and compliance with QlikView. I think using gen9 will not have critical issue, but I want to be sure if you have any surprising issue on Gen 9.

The only inportant specification on Gen 9, I think it is supported intel e5 v3 cpu family.

Any comment will be great?

HP ProLiant DL380 Gen9 Server

Thnks

Emre

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Tue, 24 Mar 2015 11:12:56 GMT

Hi Emre,

The Gen9 of HP 380 will perform very well since they upgraded the hardware to accommodate v3 of Intel Xeon E5 processors. Most of 26XX that were supported in v1 are also supported in v2 and v3, like the 2690, 2660, 2667 and 2670.

The features that did these CPUs work very well are there and have been improved (number of cores, clock cycles, cache and RAM, QPI) so I would only expect G9/v3 to perform even bette, since the basics that made them fit for QlikView (like SUMA, bridges, QPI and so) are still in place or also enhanced.

Miguel

Re: HP Gen9 Server

Tue, 24 Mar 2015 12:52:49 GMT

Hi Emre,

Use the below Blog to get an idea on Qlikview application server compatibility

http://community.qlik.com/thread/126460

Re: HP Gen9 Server

ordre — Tue, 24 Mar 2015 13:16:10 GMT

Hi Miguel,

I think the cpus below are the best fit for Gen9 and QlikView considering both clock speed and # of cores.

Intel® Xeon® Processor E5-2667 v3 (20M Cache, 3.20 GHz) (8 cores)

Intel® Xeon® Processor E5-2687W v3 (25M Cache, 3.10 GHz) (10 cores)

Thank you for prompt answer.

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Tue, 24 Mar 2015 13:30:00 GMT

Hi Emre,

I would go for the E5-2667 v3 since it is the more frequency optimized of both. It has less cores, but that means that it can take better advantage of the 9.6 GT/s QPI, that has only one ring for all the cores. The 2687W has more power consumption because of more cores, but a slower QPI ring with 8 GT/s.

If you want to use NUMA for whatever reason (QlikView is not one of them, but anyway), then go for the 2687W.

Along with some nice DDR4 2133 MHz RAM, that could be just great!

Miguel

Re: HP Gen9 Server

ordre — Tue, 24 Mar 2015 14:45:26 GMT

Hi Miguel,

Considering your choice, 6 cores Intel® Xeon® Processor E5-2643 v3 (20M Cache, 3.40 GHz) will be greater than E5-2667 v3?:)

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Tue, 24 Mar 2015 15:00:53 GMT

Emre,

I'd rather stick to the 8 cores E5-2667 v3. The QPI ring is good enough to keep all cores running and handling well.

Miguel

Re: HP Gen9 Server

Anonymous — Thu, 26 Mar 2015 14:55:08 GMT

I just posted an update my search is over!

Re: HP Gen9 Server

Anonymous — Thu, 26 Mar 2015 15:07:02 GMT

this is what we ended up getting and I am really happy with it. We also tried 4P DL580 beast and QV could not use all the power of it and it was 4 times slower than our Gen7 E5-2667.

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Thu, 26 Mar 2015 16:25:02 GMT

The HP DL580 is a beast, for QlikView you will only appreciate the performance with the E7-4870 (v1 and v2) and 4890 v2 (in case of G8) because they are part of the few only in the E7 family to have a mesh QPI bonding all cores in their CPUs therefore full access to memory modules.

Re: HP Gen9 Server

Anonymous — Thu, 26 Mar 2015 16:34:58 GMT

Miguel, this is the one we got at first

HP DL580 Gen8 E7-4870v2

but in our case it was 4 times slower (no kidding) than our good old HP DL380 Gen 7 E5-2690 2.90GHz. Granted DL580 can handle way more users than DL380 but we needed speed not capacity.

I actually hear some complaints on DL580 from our Oracle team as well - DB appears to be running slower on these boxes, but they did not do such extensive tests like i did with QlikView.

In fact I ran 4 software benchmarking tools and all 4 showed very bad numbers for DL580 in memory tests comparing to our old Gen7 DL380 - you can check my post if you are curious

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Thu, 26 Mar 2015 17:06:40 GMT

Thanks for the info BorisMan I'll be checking them indeed.

If you want speed, definitely the E5 are the ones to go with, but if you need concurrency or capacity, our tests also measured the E7s and tests were fine. You will find more detailed info in the QlikView Scalability space.

Re: HP Gen9 Server

Troy1 — Sat, 02 May 2015 05:43:04 GMT

Hi Miguel

For best speed & concurrent user performance what would you recommend out of below list.

2 socket vs 4, what should be chosen

2 socket:

E5-2697 v3 (14 cores) vs E5-2667 v3 (8 cores)

http://ark.intel.com/compare/83361,81059

any other you will recommend beside these 2.

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Mon, 04 May 2015 09:18:37 GMT

Hi Mohmed,

Between those 2, the E5-2667. The way it has the QPI ring built is more optimal for high demanding scenarios and huge memory sharing.

Apart from those, the E5-2690 is always a good choice, if not the best overall (for QlikView and Qlik Sense and in terms of performance in high resource usage environments).

Miguel

Re: HP Gen9 Server

Troy1 — Mon, 04 May 2015 12:26:16 GMT

Thanks Miguel

What do you look for in QPI configuration? what is QPI difference between E5-2667 V3 and other 2697-V3 & 2690-V3. when i compre all 3 in below link. Only difference i see is # of cores & base & turbo frequency. QPI is same for all 3.

http://ark.intel.com/compare/83361,81713,81059

E5-2667 has only 8 cores hence you will end up with 16 total. Can 1 server handle 40 to 50 concurrent users while anchoring 20gb application. Is hyperthreding allowed.

I don't see E5-2667 in Qlik list of approved server?

Re: HP Gen9 Server

Anonymous — Mon, 04 May 2015 12:51:51 GMT

our experience exactly, Miguel, I wish someone would tell us that we would save a lot of time and hassle

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Mon, 04 May 2015 13:28:45 GMT

Higher number of cores does not mean best performance when it comes to QlikView or more generally speaking, to any Uniform Memory Access apps (non-NUMA/Interleaved).

The E5-2667, being the smaller of those, has the most equitative QPI architecture, with a "ring" linking all 8 cores, 4 cores at each side (so to speak). However, the E5-2697 which has 14 cores that are unbalanced split in sides of 8 and 6, so one part of the bus will always be more loaded than the other.

Since the key for QlikView is allowing each processor to access as many memory as possible, and as even as possible, unbalancing this way could lead to eventual bottlenecks in some cores while others not so loaded.

Indeed, the E5-2667 is not in our whitelist, which does not mean it will not work, but that has not been tested in our labs. Same for the E5-2697.

Miguel

Re: HP Gen9 Server

Troy1 — Mon, 04 May 2015 14:28:26 GMT

Thanks Miguel for great insight.

You suggested E5-2690 in earlier post as best overall. Does it share the same QPI architecture as 2667.

Re: HP Gen9 Server

Miguel_Angel_Baeyens — Mon, 04 May 2015 14:42:47 GMT

The E5-2690 has proven the most powerful CPU out there for QlikView balancing both number of CPUs, available RAM and CPU speed. It does not share the same QPI configuration than the E5-2667, though.

Miguel

EDIT to my previous post: the E5-2697 v3 has been tested and is whitelisted as of Feb 2015

Re: HP Gen9 Server

Troy1 — Mon, 04 May 2015 16:52:10 GMT

Hi Miguel

E5-2690 does it have balanced QPI 6X6 as it is 12 cores? meaning faster then 2697 in terms of thruput. However would be able to handle more concurrent users as it has 4 extra cores compared to 2667.

2690 also has large smart cache compare to 2667.

2667 has advantage on higher base frequency. Turbo boost is about the same.

we have narrowed our choice to those 2 CPU