0

Im trying to determine which hardware causing the bottleneck of the server, the server is mainly used for serving video files on a heavy traffic site.

i have dstat output look like below (will get up to 500+ for send part on peak hour): enter image description here

interface output as below showing max speed is 4000Mb/s:enter image description here

below is mrtg for daily showing 500Mbyets max speed:enter image description here

The hardisk info using: smartctl -a /dev/sda

HDD info:
Vendor:               DELL
Product:              PERC H710P
Revision:             3.13
User Capacity:        1,999,307,276,288 bytes [1.99 TB]
Logical block size:   512 bytes
Logical Unit id:      0x6b82a720d22304002116d6c01027fc4d
Serial number:        004dfc2710c0d61621000423d220a782
Device type:          disk
Local Time is:        Mon Sep 11 09:06:33 2017 CST
Device does not support SMART

Error Counter logging not supported
Device does not support Self Test logging

how do i know which one is causing the bottleneck? using pure-ftp i always get listing directory failed, once i disable the nginx, i immediately able to list the directory. Now im not sure which is causing the bottleneck problem, either the max hdd data read or the network bandwidth, please help me so that i can decide to add a new hdd or add bandwidth.

Teddybugs
  • 153
  • 10
  • A single spinning hard disk can do at most 200MB/sec (1600Mbps), usually less, and accessing multiple files it will have a lower total throughput. An SSD can probably do three or four times that, with no slow down. What kind of disk(s) do you currently have? – Tim Sep 11 '17 at 01:03
  • @Tim : updated disk info on question, not sure it is ssd or hdd, how do i check. – Teddybugs Sep 11 '17 at 01:10
  • lsblk -d -o name,rota NAME ROTA sda 1 that mean it is hdd? – Teddybugs Sep 11 '17 at 01:13
  • Sorry but I don't have the knowledge to help you work that out. If it's a physical server you can look at it, if it's hosted you can ask your provider or look at the bill. A Linux guru could tell you what command to use to work it out. – Tim Sep 11 '17 at 01:28

1 Answers1

1

Based on provided information, I guess the problem is in the network speed Your peak at ~500 MB is the same as maximum throughput of your bonded network interface (4000 Mb).

The information about the disk says that it's Dell RAID controller. So real disks are hidden and the disk controller gives you virtual drive based on some type of RAID array. We use HPE hardware so I can't give you any command for DELL array which would give you more information. However I believe you can google it easily or someone please attach it as comment.

Jaroslav Kucera
  • 1,545
  • 11
  • 18
  • i have checked the dell idrac interface, it showing 3x900GB~ ssd, btw how do you know/calculate the maximum throughput of network interface 4000 Mb is actually =~ 500MB, do you have specific formula for that? might need it for calculate other server as well. – Teddybugs Sep 11 '17 at 08:14
  • got my answer about calculation here: https://support.myrepublic.com.sg/hc/en-us/articles/202898774-How-fast-is-1Gbps-fibre-broadband- 4000Mb = 125MBx4 = 500MB~ – Teddybugs Sep 11 '17 at 08:27
  • That data above shows that connection rates are practically limited to a total of 500 megaBITS, not megabytes. I noticed that there isn't any port negotiation on your bond. Are you confident that traffic is going down all links? Also, what kind of bond is this? If it's a balance-rr rather than a 802.3ad bond, you'll get horrible performance on most protocols. – Spooler Sep 11 '17 at 08:32
  • @SmallLoanOf1M You can get horrible performance with LACP trunks as well - it all depends on the load-balancing setup and algorithm which isn't standardized. – Zac67 Sep 11 '17 at 11:15