Navneet Kumar's Library tagged Hardware → View Popular

Filter:
All
Annotated

29 Sep 09

Facebook | Needle in a haystack: efficient storage of billions of photos

READ WRITE Performance

www.facebook.com/note.php - Preview

FaceBook Infrastructure Architecture Hardware Performance Read-Write

Since each image is stored in its own file, there is an enormous amount of
metadata generated on the storage tier due to the namespace directories and file
inodes. The amount of metadata far exceeds the caching abilities of the NFS
storage tier, resulting in multiple I/O operations per photo upload or read
request. The whole photo serving infrastructure is bottlenecked on the high
metadata overhead of the NFS storage tier, which is one of the reasons why
Facebook relies heavily on CDNs to serve photos
RAID-6 partition managed by the hardware RAID controller. RAID-6 provides
adequate redundancy and excellent read performance while keeping the storage
cost down
1 more annotations...
- The poor write performance is partially mitigated by the RAID controller NVRAM
  write-back cache. Since the reads are mostly random, the NVRAM cache is fully
  reserved for writes. The disk caches are disabled in order to guarantee data
  consistency in the event of a crash or a power loss

25 Sep 09

RAID - Wikipedia, the free encyclopedia

en.wikipedia.org/RAID - Preview

Disk Hardware RAID

RAID can involve significant computation when reading and writing information.
With traditional "real" RAID hardware, a separate controller does this
computation. In other cases the operating system or simpler and less expensive
controllers require the host computer's processor to do the computing, which
reduces the computer's performance on processor-intensive tasks
RAID's main aim can be either to improve reliability and availability of data
8 more annotations...
- Basic mirroring can speed up reading data as a system can read different data
  from both the disks, but it may be slow for writing if the configuration
  requires that both disks must confirm that the data is correctly written.
  Striping is often used for performance, where it allows sequences of data to be
  read from multiple disks at the same time. Error checking typically will slow
  the system down as data needs to be read from several places and compared
- The design of RAID systems is therefore a compromise and understanding the
  requirements of a system is important
- The software must run on a host server attached to storage, and server's
  processor must dedicate processing time to run the RAID software. This is
  negligible for RAID 0 and RAID 1, but may become significant when using
  parity-based arrays and either accessing several arrays at the same time or
  running many disks. Furthermore all the busses between the processor and the
  disk controller must carry the extra data required by RAID which may cause
  congestion
- For data safety in modern systems the user of software RAID might need to turn
  the write-back cache on the disk off
- Turning off the write cache has a performance penalty that can, depending on
  workload and how well supported command queuing in the disk system is, be
  significant. The battery backed cache on a RAID controller is one solution to
  have a safe write-back cache
- Hardware RAID controllers use different, proprietary disk layouts, so it is not
  usually possible to span controllers from different manufacturers. They do not
  require processor resources, the BIOS can boot from them, and tighter
  integration with the device driver may offer better error handling
- In most systems the write cache is non-volatile (i.e. battery-protected), so
  pending writes are not lost on a power failure.
- Most hardware implementations provide a read/write cache, which, depending on the I/O workload, will improve
  performance

15 Jun 09

What Your Computer Does While You Wait : Gustavo Duarte

CPU - Cache - Memory - Disk

duartes.org/...r-computer-does-while-you-wait - Preview

Performance Computer-Architecture Computing Hardware Windows

one clock cycle to execute, hence a third of a nanosecond at 3.0Ghz. For reference, light only travels ~4 inches (10 cm) in the time taken by a clock cycle.
To put this into perspective, reading from L1 cache is like grabbing a piece of paper from your desk (3 seconds), L2 cache is picking up a book from a nearby shelf (14 seconds), and main system memory is taking a 4-minute walk down the hall to buy a Twix bar
3 more annotations...
- For a discussion of all things memory, see Ulrich Drepper’s What Every Programmer Should Know About Memory (pdf), a fine paper on the subject.
- Keeping with the office analogy, waiting for a hard drive seek is like leaving the building to roam the earth for one year and three months.
- Filesystem defragmentation aims to keep files in continuous chunks on the disk to minimize seeks and boost throughput.