Big performance gains in Wordpress from storage update?

[Patrick]

Here is a crazy one for you (and @William you can try this in the back end to verify.) I think the new STH WP VM is significantly faster than the old one in terms of one key metric: WordPress image upload speed.

Old server
KVM 8 core VM on a dual E5-2690 V1 system with 128GB RAM
2x SanDisk LB406s 400GB LSI SAS2 controller ZFS mirror

SanDisk drives were dedicated to STH WordPress so nothing else was running on this machine.

New server
KVM 8-core VM on a dual E5-2699 V3 system with 128GB RAM
2x Intel DC P3600 400GB

Intel SSDs dedicated to the STH WordPress VM.

Bandwidth between the servers and home should be about the same 95-100mbps. Ping differential is "large" - 21ms to Fremont and 25ms to Las Vegas.

Here is the strange part - the new setup seems to be uploading and processing images in WordPress significantly faster than the old one. In the realm of 60-70% faster on bulk uploads. Enough that it is easily tangible.

The E5 V3 is faster clock for clock but the E5 V1 has much higher clock speeds.

I wonder how I can turn this into a benchmark if I am not seeing things.

 

[OBasel]

How close are the settings on each VM?

 

[T_Minus]

Same WP plugins / image management in WP on the new VM?

 

[Patrick]

@OBasel and @T_Minus - same ZFS snapshot.

 

[William]

Yes the V3's really perform nice and like you said its a huge dif vs the older V1's

 

[T_Minus]

SanDisk Lightning 406s 400GB SLC SAS SSD Benchmarks

Maybe those old drives were that slow?

 

[Patrick]

SanDisk Lightning 406s 400GB SLC SAS SSD Benchmarks

Maybe those old drives were that slow?

Well the stranger thing is that it is very perceptible uploading 10-15 images 5MB total. That should be a transfer small enough where it would not matter much.

 

[mstone]

Well the stranger thing is that it is very perceptible uploading 10-15 images 5MB total. That should be a transfer small enough where it would not matter much.

Maybe the wordpress code really is that bad...

 

[TuxDude]

Have you watched the perf counters on both servers to see where it is bottlenecked? I usually look at things like this from a "it would finish instantly if it wasn't waiting for something" point-of-view, and then go about figuring out what the thing being waited on is.