Server Fundamentals Flashcards
Print Server?
Manages print request sent by the network
Mail Server?
Manages the transmission of email
Web Server?
Houses a website and manages requests to view that site
File Server?
Provides centralized file storage that can be accessed over the network
Database Server?
Stores and provides access to a database of information
Application Server?
Dedicated to running certain software applications
Tower Server Pros
Traditional Form Factor
Provides the most internal expandability
Includes all necessary components
Tower Server Cons
Takes up a lot of space
Difficult to add additional servers
Rack Servers
Density (more CPUs and RAM in less space)
Easy to add additional servers
Rack enclosur offers neater cabling
How much is a U?
1.75in.
What are the most common rack sizes?
42U and 48U
PCIe Expansion Slots?
allow for additional and faster network ports
Management Port?
for remote configuration and server management
Relation between CPUs and Power?
More CPUs = More Power needed
Why servers have multiple power supplys?
Redundancy
are hot swappable
Modular Servers?
can support 8 full height / 16 half height / 32 quarter height servers
- 16 half height servers are 60% less dense than 1U rack server
Enclosure houses all power, cooling, networking, and server to server communication (minimizes cabling)
What is point of RAID controllers?
remove the load on the processors
in case of drive failure controller can rebuild data on a “hot spare”
Quoting Chassis?
10 drive or 8 drive chassis
- can’t be changed after point of sale / can’t upgrade after
Motherboard?
The glue
- CPUs plug into sockets
- everything on motherboard ties back to CPUs
- copper lanes tie back to CPUs
- more lanes = more bandwidth
Quoting Processor?
aka CPU
- Intel Xeon or AMD EPYC
- Xeon has no integrated graphics card
- 2.3 GHz Speed -> each core will run at that speed
Quoting Memory?
aka RAM
- applications run in RAM
- measured on DIMM cards -> measured by GBs 8-128gbs
- Sweet Spot = Quad 32 and 64 gbs
What are PSUs?
Power Supply Unit
- device in server that regulates power delivery
- within our server we sell different wattages (495W-3000W) (size power supplies according to wattage)
- Hot swappable but must shut down server to change out
NICs?
cards that give connectivity to the network
- Goes in PCIe in the back
- Has its own CPU
- must come from us
Host Bus Adapter?
special type of card for storage traffic and communication
- PCIe slot
BOSS Card?
Boot optimized storage solution
- Separates the server OS from physical drive
- provides performance benefits and redundancy w/ RAID1
- 240gb or 480gb sizes
PERC?
PowerEdge RAID Controller
- piece of hardware that controls the RAID
- H330 = no cache or RAID 5/6
- H730P = 2GB Cache (Standard)
- H740P = 8GB Cache
Cache?
PERC - Stores in bucket AKA Cache Temporarily to increase performance and acknowledge back to app quicker.
Controller recognizes patterns of information and starts loading info into cache.
iDRAC
How we manage our servers remotely
- can enable lockdown and set up support assist remotely
- monitor power usage and component health
Riser Card
plugs into motherboard and cards can stack horizontally
- different bandwidths on PCIe slots (by8/by16)
- slots on PCIe tell you what size by and what CPU is associated with each
Mezz Card
Networking card for blade servers
PCIe Slots
peripheral component interconnect express
- gives you flexibility
- expansion slots for network cards
GPU
general purpose graphic processing unit
- for parrallel processing in a data center
- CPU does one task at a time GPU does multiple at same time
- server must have right hardware due to power level and heat
- must make sure server is GPU compatible (can’t use consumer GPUs)
NVMe PCIe
NVMe plugs into PCIe
- gives you super fast storage (1,000,000 IOPS)
- fast due to protocol and no RAID controller
- connected directly to CPU
PowerEdge Form Factor
R740
- R = form factor (Rack)
-7 = lvl of capability
4 = generation
0 = processor type (Intel)
R7515
- R = form factor (Rack)
-7 = lvl of capability
5 = generation
1 = # of processors
5 = processor type (AMD)
Drive Max?
1U = 10 drives
2U = 32 drives
PMSN For Server Qualification?
Processor
Memory
Storage
NIC
FIle Server Requirements?
low processing power + low memory
higher storage + broad network access
Backup Server Requirements?
low processing power + low memory
higher storage + broad network access
What servers require high everything for PMSN?
Email and Database Servers
What server requires low everything for PMSN?
domain controller
Most important parts of virtual host sizing?
processing power + memory
- shared storage so don’t need a bunch
Hardware RAID?
PERC - controller w/ integrated CPU handles all of the calculations
Software RAID?
does not use RAID controller - operates at the OS level
- CPU/Server is doing the RAID work
- could cause performance bottlenecks for OS since it’s focused more on RAID
IOPS?
Inputs/Outputs per second
- very dependent on drive types
Throughput?
Size of the “lane” between the server’s hard drive controller and the disk
- most common speed is 12Gb/s
Random vs Sequential
random = found in most corporate apps (Exchange/SQL) + most taxing on drives
sequential = less common/less taxing - typically video streaming and files
Rules of Thumb on IOPS per Drive
7.2K SATA = 80 IOPS
10K SAS = 120 IOPS
15K SAS = 180 IOPS
SSDs = 2500-10,000 IOPS
RAID 0
solves performance challenges by striping the data across two or more hard drives.
- All of the drives resources are combined and linearly increase performance as drives are added to the RAID set
- requires 2 drive minimum + no redundancy/failure tolerance
- usable capacity of all drives combined
RAID 1
creates a mirror of one drive to a second drive.
- This helps solve data protection challenges, as one drive can fail and the other drive can keep the server running as it has a copy of all of the data.
- not expandable (2 drive min/max and limited to capacity of 1 drive)
- used a lot for OS
- limited to one drives IOPS for writes but can use both drives IOPS for reads
RAID 10
striping + mirroring
- This leverages the strength of RAID 1 with its redundancy with the performance of RAID 0
- RAID 10 is used for applications that have a high random write data pattern, as it does not have to use parity
- RAID 10 does not provide as good of a capacity utilization, as half of the drives capacity are lost to the mirrors.
RAID 5
striping w/ single parity
- Number of hard drives minus 1 for parity, need at least 3
- As long as this parity data is present, if any piece of data is lost, it can be recovered by putting all of the remain piece of data and the parity data back into the equation.
- efficient balance of performance/capacity (66-92% efficiency)
- used a lot for general purpose applications with high reads like Exchange
- fast for reads but slow for writes and can expand one drive at a time
RAID 6
Striping + Double Parity
- 2 drives can be lost without losing data + Loss of capacity because of this. (50-90% efficiency)
- Slow write time
- Use: Backup, file saving
- Fast on reads, slow on writing (4 drive min.)
- helpful where multi-TB disks are necessary (take a longer time to rebuild
CPUs
brains of the server
- made up of one or more cores (each core is a processor)
- more cores = more things done at once
CPU Clock Speed
the rate at which a CPU can process instructions (aka BIN Speed)
- measured in GHz
- Turbo Boost = accelerates performance for peak loads to allow cores to run faster that rated frequency for short bursts
Hyperthreading
enables multiple sets of instructions to run on each processor core (improves overall peformance)
Intel Processor Nomenclature
Intel Xeon Platinum 8280M
- Intel Xeon = brand
- Platinum = brand modifier
- 8 = SKU Level (8(platinum),6/5(gold),4(silver),3(bronze)
- 2 = processor generation
-80 = processor SKU - M = memory capacity (no suffix(1TB per socket),M(2 TB per socket), L(4.5TB per socket)
Intel Processor Integrations and Optimizations
N = NFV Optimized
S = Search Optimized
T = High Tcase/extended reliability
U = single socket
V = VM density optimized
Y = Intel Speed Tech Technology
Importance of processor Wattage?
will determine the maximum capacity of the processors that can be configured for a system.
most two-processor configurations will only support 1.5TB of memory.
RDIMM?
registered DIMM
- manages memory bandwidth based on a register chip (keeps all I/Os in line)
- loading factor increases w/ frequency - memory becomes slower as the register chip must manage more and more memory channels
- 8-64Gbs
LRDIMM
Load Reduced DIMM
- technology that uses buffer chip to assist register chip
- allows bandwidth to be distributed more evenly (minimized loading factor at high memory densities) - enables higher capacity and performance per CPU/server
- Quad rank LRDIMMs are treated as dual rank allowing for more 128GB LRDIMMs per channel
- 128Gbs
How much DIMM on Intel vs AMD?
Intel = 6 or 12
AMD = 8 or 16
NVDIMM-N
combines DRAM and NAND flash on a single DIMM module
- behaves like RDIMM in normal operation
- there is no direct access to the flash memory on the DIMM
- uses multiplexers instead of the data buffer found on LRDIMM
- in the event of power failure the memory contents are saved to flash memory
Memory Rules
Mixing of different memory types in servers is restricted
- ability to mix RDIMM and NVDIMM is supported with the restriction of only one NVDIMM per channel
- must populate all slots in memory for best performance “balanced memory config” (8 or 16 for AMD and 6 or 12 for Intel PER CPU
Memory Ranking
DIMM Module is made up of DRAM chips
- when data is written to memory it’s saved on DRAM chips
- DRAM chips are grouped into a rank
- larger the memory the more ranks there are
- chips can be either single, dual, quad or octal
iDRAC competitors
HPE = iLO
Lenovo = XCC
Supermicro = SSM
