Hello to my favorite Epicor group! So we are working with a consultant ramping up to our E10 upgrade. We have an all new virtualized environment with very very decent hardware. Please don’t ask me the details of it because I truly was not involved in it as I am an Epicor developer. Anyhow, our consultant was checking performance prior to installing SQL and Epicor and such. They are saying that we HAVE to hit a latency of a maximum of 5 milliseconds! They say this is a requirement of Epicor 10. This sounds absolutely ridiculous to us without spending $100k on all new hardware.
Can anyone shed some light on this for me? Any experience with this? What is your latency and how is your E10 performance?
A latency on what aspect?
There is network latency, transmision, propagation, wan, audio, io, operating latency between input and command execution, there’s a lotta “latency”… sounds to me like your consultant is blowing smoke …
How many users? Db on same server as AppServer? and yes we need the specs… (i know you said don’t ask but we can’t help if we don’t know what you are working with)
Thank you. I’m getting those details. Separate app server and DB server. Yes on San. The latency they are speaking up is the disc IO on the DB server cannot be more than 5 ms.
First thing first. Are you deploying Epicor on virtual servers? If not, a single box can do the trick and you don’t need to separate the application server and DB server.
Secondo, a SAN is not required for small or medium deployment. Local SSD (flash) disks will be sufficient.
Also, consultant recommended 3 VMs for Epicor with the 3rd as a tools server, a work space for them during the upgrade and I believe this one may run service connect.
30 concurrent and 10 data collection. I would say the load is between small and medium.
Just so you know, 1 host is enough to manage the load. You don’t buy two hosts to get more performance. The reason why you want at least 2 host is you can manage failover. So if anything goes wrong with the first host and it goes down, virtual machines running on the first host will failover the second host, so you don’t have any down time.
As for the disc latency, what I am assuming so correct me if I’m wrong, is that this is recommended by Epicor so that it runs optimal but not required. Keeps us that don’t have the deep pockets from calling and complaining about performance problems. Last week, I installed and I ran the client from my workstation using the Train environment. Compared to our 9.05, it was screaming fast for me. Granted, I was the only client connected.
The SAN will not be a problem. The only this is having everything on RAID 10, you loose a lot of disk space. Databases and log files should be on RAID 10, while RAID 1 is enough for the remaining (like OS, etc.).
It’s not out of the ordinary for the vendor to ask for sub 5 millisecond ranges. I’ve seen this on the smallest systems with the most basic requirements. But they’re leaving out a whole host of questions - like what kind of data access does that mean, and in what storage context?
We just deployed in a fully virtualized environment, including SQL for the first time. Our SAN environment is nowhere in the EMC range, but our user count is above yours, and we use a single VM host btw.
What you absolutely should consider is that it’s folly to focus in strictly on Diskspd results without understanding more about SAN queuing, and if your SAN vendor has special setup scenarios for SQL Server (many do).
We ended up placing our SQL on a fully isolated SAN storage area, and the logs in SAN SSD. We worked with the SAN vendor to run tests (the usual ones you can find for SQL such as Diskspd) and understand what we were seeing. I really suggest your DBA/Networking/SysAdmin teams work together to achieve a baseline of performance before things are stood up.
There’s a lot of literature out there about trying to reconcile what you would see with direct attached storage vs SAN.
A few ideas:
Check out HammerDB
Check out the VMWare white paper on SQL Server on VMWare best practices. Your sysadmin team needs to know about this because it’s definitely not like just standing up another random server.
how big is your DB? slap it on an intel P4800x get VMware to replicate to a second server also running at P4800x and you’d be talking maybe $20K of hardware/licensing