[SOLVED] 4 SCSI drives for OS in RAID, 1 SSD for data: good config ?

[eugenio.bartolo]

Hi Everyone,

A client of mine has been provided a new server with 4 Toshiba SCSI drives in RAID (configuration should be RAID-5).
Then they have an SSD where they store data that are constantly used (in a CRM-like system).

The client is experiencing a very annoying slow performance of the CRM.

Here is the specs of the server:

• Processor: Intel E5-2609V4 Octa Core (1.7 GHZ)
  Chipset: Intel C612
  OS: Windows Server 2016 64bit
  Drives: 4 Toshiba SCSI 10000rpm 300Gb; 1 Samsung SSD 240Gb

Further details available here.

Do you think having the data in a SSD drive, which is not part of the RAID array, creates a bottleneck to performances ?

Thanks for your help !

 

[drtweak]

I hope you mean SAS and not SCSI.

I don't see where there would be a bottle neck anywhere if they are actually SAS drives. If they really as 80 pin SCSI drives I have to first ask...WTH are you guys using 15 year old drives for? lol

The first thing i would try is during the slow down, run the CRM on the server. is it slow then? If so something wrong with server. if it is now slow then I would actually check the network next.

 

[eugenio.bartolo]

Hey @drtweak thanks for your reply !

Ha - just to confirm, it's SAS drives.
I'll see what happens with running the CRM on the server, on the SAS drives, rather than on the SSD.

 

[drtweak]

Hey @drtweak thanks for your reply !

Ha - just to confirm, it's SAS drives.
I'll see what happens with running the CRM on the server, on the SAS drives, rather than on the SSD.

Well you weren't Totally wrong by saying SCSI though because SAS means Serial Attached SCSI XD

But yea during the slow down runt he CRM directly off the server and see if it is slow as well. Also check things like RAM, CPU, and Network usage during slowdown on the server as well.

 

[Solandri]

How full is the SSD and how frequently is data written to it?

On a HDD, you can overwrite a 1 with a 0, or a 0 with a 1. SSDs can't do that. They first have to erase the memory cell before new data can be written to it (it's like the old EEPROMs). That erase step is painfully slow. Normally SSDs will erase sectors with deleted data during idle time (which is why TRIM was important). That way there's always a bunch of pre-erased sectors ready and available for lightning-fast writes. But if the SSD is very full, or data is being written to it faster than it can erase deleted sectors, the SSD's performance will plummet to worse-than-HDD speeds. Even read speeds will suffer as the SSD will be so busy erasing that it cannot handle read requests.

Also, enterprise operations which write a lot of data should go with SLC or MLC SSDs. Avoid the TLC and QLC models. Writes to TLC or QLC NAND are a lot slower. So what those SSDs do is initially write data in SLC mode. Then later during idle time, it will rewrite that data in TLC/QLC mode. But if you're writing too much data, the SLC buffer will fill up. The SSD will be forced to write directly in TLC/QLC mode, which will slow it down considerably.