Friday, March 26, 2010

Understanding /proc/cpuinfo?

A hyperthreaded processor has the same number of function units as an older, non-hyperthreaded processor. It just has two execution contexts, so it can maybe achieve better function unit utilization by letting more than one program execute concurrently. On the other hand, if you're running two programs which compete for the same function units, there is no advantage at all to having both running "concurrently." When one is running, the other is necessarily waiting on the same function units.

A dual core processor literally has two times as many function units as a single-core processor, and can really run two programs concurrently, with no competition for function units.

A dual core processor is built so that both cores share the same level 2 cache. A dual processor (separate physical cpus) system differs in that each cpu will have its own level 2 cache. This may sound like an advantage, and in some situations it can be but in many cases new research and testing shows that the shared cache can be faster when the cpus are sharing the same or very similar tasks.

In general Hyperthreading is considered older technology and is no longer supported in newer cpus. Hyperthreading can provide a marginal (10%) for some server workloads like mysql, but dual core technology has essentially replaced hyperthreading in newer systems.

A dual core cpu running at 3.0Ghz should be faster then a dual cpu (separate core) system running at 3.0Ghz due to the ability to share the cache at higher bus speeds.

The examples below details how we determine what kind of cpu(s) are present.

The kernel data Linux exposes in /proc/cpuinfo will show each logical cpu with a unique processor number. A logical cpu can be a hyperthreading sibling, a shared core in a dual or quad core, or a separate physical cpu. We must look at the siblings, cpu cores and core id to tell the difference.

If the number of cores = the number of siblings for a given physical processor, then hyperthreading is OFF.

/bin/cat /proc/cpuinfo | /bin/egrep 'processor|model name|cache size|core|sibling|physical'

Example 1: Single processor, 1 core, no Hyperthreading

processor : 0
model name : AMD Duron(tm) processor
cache size : 64 KB

Example 2: Single processor, 1 core, Hyperthreading is enabled.

Notice how we have 2 siblings, but only 1 core. The physical cpu id is the same for both: 0.

processor : 0
model name : Intel(R) Pentium(R) 4 CPU 2.80GHz
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
processor : 1
model name : Intel(R) Pentium(R) 4 CPU 2.80GHz
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1


Example 3. Single socket Quad Core

Notice how each processor has its own core id. The number of siblings matches the number of cores so there are no Hyperthreading siblings. Also notice the huge l2 cache - 6 MB. That makes sense though, when considering 4 cores share that l2 cache.

processor : 0
model name : Intel(R) Xeon(R) CPU E5410 @ 2.33GHz
cache size : 6144 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 4
processor : 1
model name : Intel(R) Xeon(R) CPU E5410 @ 2.33GHz
cache size : 6144 KB
physical id : 0
siblings : 4
core id : 1
cpu cores : 4
processor : 2
model name : Intel(R) Xeon(R) CPU E5410 @ 2.33GHz
cache size : 6144 KB
physical id : 0
siblings : 4
core id : 2
cpu cores : 4
processor : 3
model name : Intel(R) Xeon(R) CPU E5410 @ 2.33GHz
cache size : 6144 KB
physical id : 0
siblings : 4
core id : 3
cpu cores : 4

Example 3a. Single socket Dual Core

Again, each processor has its own core so this is a dual core system.

processor : 0
model name : Intel(R) Pentium(R) D CPU 3.00GHz
cache size : 2048 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
processor : 1
model name : Intel(R) Pentium(R) D CPU 3.00GHz
cache size : 2048 KB
physical id : 0
siblings : 2
core id : 1
cpu cores : 2

Example 4. Dual Single core CPU, Hyperthreading ENABLED

This example shows that processer 0 and 2 share the same physical cpu and 1 and 3 share the same physical cpu. The number of siblings is twice the number of cores, which is another clue that this is a system with hyperthreading enabled.

processor : 0
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
processor : 1
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cache size : 1024 KB
physical id : 3
siblings : 2
core id : 0
cpu cores : 1
processor : 2
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cache size : 1024 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 1
processor : 3
model name : Intel(R) Xeon(TM) CPU 3.60GHz
cache size : 1024 KB
physical id : 3
siblings : 2
core id : 0
cpu cores : 1

Example 5. Dual CPU Dual Core No hyperthreading

Of the 5 examples this should be the most capable system processor-wise. There are a total of 4 cores; 2 cores in 2 separate socketed physical cpus. Each core shares the 4MB cache with its sibling core. The higher clock rate (3.0 Ghz vs 2.3Ghz) should offer slightly better performance than example 3.

processor : 0
model name : Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
cache size : 4096 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
processor : 1
model name : Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
cache size : 4096 KB
physical id : 0
siblings : 2
core id : 1
cpu cores : 2
processor : 2
model name : Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
cache size : 4096 KB
physical id : 3
siblings : 2
core id : 0
cpu cores : 2
processor : 3
model name : Intel(R) Xeon(R) CPU 5160 @ 3.00GHz
cache size : 4096 KB
physical id : 3
siblings : 2
core id : 1
cpu cores : 2

7 comments:

  1. Nice article, made things clear.

    --Chithu

    ReplyDelete
  2. great article however things are still unclear for me :)

    have to read again

    Http://www.nixsup.com

    ReplyDelete
  3. If this was you tube i'd Like and Favourite! Well explained that man!

    ReplyDelete
  4. thanks, read again, finally I got it

    ReplyDelete
  5. "Hyperthreading can provide a marginal (10%) for some server workloads ", that seems hyperthreading can only provide a small marginal . Then why when hyperthreading ENABLED,the cpuinfo would show twice processor items, which would give a fake impression of twice process power .

    ReplyDelete
    Replies
    1. While for some servers workload the improvement can be marginal, it can be drastic with other kinds of workload (ex: heavy IO wait workloads). /proc/cpuinfo entries are execution contexts, hyper-threading allows each core to maintain two execution contexts (or threads), so it doubles the number of cpuinfo entries. Hyperthreading allows a core to switch quickly to the other execution context when the current one is starving for IO (from network, disks or RAM...) or when it sleeps, without having to replace the entire instruction pipeline, and allows to switchback fast too when first thread is ready to work again.

      Delete