The MESI protocol is an Invalidate-based cache coherence protocol, and is one of the most common protocols that support write-back caches. It is also known as the Illinois protocol due to its development at the University of Illinois at Urbana-Champaign.[1] Write back caches can save considerable bandwidth generally wasted on a write through cache. There is always a dirty state present in write-back caches that indicates that the data in the cache is different from that in the main memory. The Illinois Protocol requires a cache-to-cache transfer on a miss if the block resides in another cache. This protocol reduces the number of main memory transactions with respect to the MSI protocol. This marks a significant improvement in performance.[2]
The letters in the acronym MESI represent four exclusive states that a cache line can be marked with (encoded using two additional bits):
For any given pair of caches, the permitted states of a given cache line are as follows:
When the block is marked M (modified) or E (exclusive), the copies of the block in other Caches are marked as I (Invalid).The MESI protocol is defined by a finite-state machine that transitions from one state to another based on 2 stimuli.
The first stimulus is the processor-specific Read and Write request. For example: A processor P1 has a Block X in its Cache, and there is a request from the processor to read or write from that block.
The second stimulus is given through the bus connecting the processors. In particular, the "Bus side requests" come from other processors that don't have the cache block or the updated data in their Cache. The bus requests are monitored with the help of Snoopers,[3] which monitor all the bus transactions.
Following are the different types of Processor requests and Bus side requests:
Processor Requests to Cache include the following operations:
Bus side requests are the following:
(Such Cache to Cache transfers can reduce the read miss latency if the latency to bring the block from the main memory is more than from Cache to Cache transfers, which is generally the case in bus based systems.)
Snooping Operation: In a snooping system, all caches on a bus monitor all the transactions on that bus. Every cache has a copy of the sharing status of every block of physical memory it has stored. The state of the block is changed according to the State Diagram of the protocol used. (Refer image above for MESI state diagram). The bus has snoopers on both sides:
Explanation:
Each Cache block has its own 4 state finite-state machine (refer image 1.1). The State transitions and the responses at a particular state with respect to different inputs are shown in Table1.1 and Table 1.2
| Invalid(I) | PrRd |
| |
| PrWr |
| ||
| Exclusive(E) | PrRd |
| |
| PrWr |
| ||
| Shared(S) | PrRd |
| |
| PrWr |
| ||
| Modified(M) | PrRd |
| |
| PrWr |
|
| Invalid(I) | BusRd |
| |
| BusRdX/BusUpgr |
| ||
| Exclusive(E) | BusRd |
| |
| BusRdX |
| ||
| Shared(S) | BusRd |
| |
| BusRdX/BusUpgr |
| ||
| Modified(M) | BusRd |
| |
| BusRdX |
|
A write may only be performed freely if the cache line is in the Modified or Exclusive state. If it is in the Shared state, all other cached copies must be invalidated first. This is typically done by a broadcast operation known as Request For Ownership (RFO).
A cache that holds a line in the Modified state must snoop (intercept) all attempted reads (from all the other caches in the system) of the corresponding main memory location and insert the data that it holds. This can be done by forcing the read to back off (i.e. retry later), then writing the data to main memory and changing the cache line to the Shared state. It can also be done by sending data from Modified cache to the cache performing the read. Note, snooping only required for read misses (protocol ensures that Modified cannot exist if any other cache can perform a read hit).
A cache that holds a line in the Shared state must listen for invalidate or request-for-ownership broadcasts from other caches, and discard the line (by moving it into Invalid state) on a match.
The Modified and Exclusive states are always precise: i.e. they match the true cache line ownership situation in the system. The Shared state may be imprecise: if another cache discards a Shared line, this cache may become the sole owner of that cache line, but it will not be promoted to Exclusive state. Other caches do not broadcast notices when they discard cache lines, and this cache could not use such notifications without maintaining a count of the number of shared copies.
In that sense the Exclusive state is an opportunistic optimization: If the CPU wants to modify a cache line in state S, a bus transaction is necessary to invalidate all other cached copies. State E enables modifying a cache line with no bus transaction.
Illustration of MESI protocol operations
Let us assume that the following stream of read/write references. All the references are to the same location and the digit refers to the processor issuing the reference.
The stream is : R1, W1, R3, W3, R1, R3, R2.
Initially it is assumed that all the caches are empty.
Note: The term snooping referred to below is a protocol for maintaining cache coherency in symmetric multiprocessing environments. All the caches on the bus monitor (snoop) the bus if they have a copy of the block of data that is requested on the bus.A Read For Ownership (RFO) is an operation in cache coherency protocols that combines a read and an invalidate broadcast. The operation is issued by a processor trying to write into a cache line that is in the shared (S) or invalid (I) states of the MESI protocol. The operation causes all other caches to set the state of such a line to I. A read for ownership transaction is a read operation with intent to write to that memory address. Therefore, this operation is exclusive. It brings data to the cache and invalidates all other processor caches that hold this memory line. This is termed "BusRdX" in tables above.
MESI in its naive, straightforward implementation exhibits two particular performance issues. First, when writing to an invalid cache line, there is a long delay while the line is fetched from other CPUs. Second, moving cache lines to the invalid state is time-consuming. To mitigate these delays, CPUs implement store buffers and invalidate queues.[4]
A store buffer is used when writing to an invalid cache line. As the write will proceed anyway, the CPU issues a read-invalid message (hence the cache line in question and all other CPUs' cache lines that store that memory address are invalidated) and then pushes the write into the store buffer, to be executed when the cache line finally arrives in the cache.
A direct consequence of the store buffer's existence is that when a CPU commits a write, that write is not immediately written in the cache. Therefore, whenever a CPU needs to read a cache line, it first scans its own store buffer for the existence of the same line, as there is a possibility that the same line was written by the same CPU before but hasn't yet been written in the cache (the preceding write is still waiting in the store buffer). Note that while a CPU can read its own previous writes in its store buffer, other CPUs cannot see those writes until they are flushed to the cache - a CPU cannot scan the store buffer of other CPUs.
With regard to invalidation messages, CPUs implement invalidate queues, whereby incoming invalidate requests are instantly acknowledged but not immediately acted upon. Instead, invalidation messages simply enter an invalidation queue and their processing occurs as soon as possible (but not necessarily instantly). Consequently, a CPU can be oblivious to the fact that a cache line in its cache is actually invalid, as the invalidation queue contains invalidations that have been received but haven't yet been applied. Note that, unlike the store buffer, the CPU can't scan the invalidation queue, as that CPU and the invalidation queue are physically located on opposite sides of the cache.
As a result, memory barriers are required. A store barrier will flush the store buffer, ensuring all writes have been applied to that CPU's cache. A read barrier will flush the invalidation queue, thus ensuring that all writes by other CPUs become visible to the flushing CPU. Furthermore, memory management units do not scan the store buffer, causing similar problems. This effect is visible even in single threaded processors.[5]
The most striking difference between MESI and MSI is the extra "exclusive" state present in the MESI protocol. This extra state was added as it has many advantages. When a processor needs to read a block that none of the other processors have and then write to it, two bus transactions will take place in the case of MSI. First, a BusRdX request is issued to read the block followed by a BusUpgr request before writing to the block. The BusRdX request in this scenario is useless as none of the other caches have the same block, but there is no way for one cache to know about this. Thus, MESI protocol overcomes this limitation by adding an Exclusive state, which results in saving a bus request. This makes a huge difference when a sequential application is running. As only one processor works on a piece of data, all the accesses will be exclusive. MSI performs much worse in this case due to the extra bus messages. Even in the case of a highly parallel application with minimal sharing of data, MESI is far faster. Adding the Exclusive state also comes at no cost as 3 states and 4 states are both representable with 2 bits.
In case continuous read and write operations are performed by various caches on a particular block, the data has to be flushed to the bus every time. Thus, the main memory will pull this on every flush and remain in a clean state. But this is not a requirement and is just an additional overhead caused by using MESI. This challenge was overcome by the MOESI protocol.[6]
In case of S (Shared State), multiple snoopers may response with FlushOpt with the same data (see the example above). The F state in MESIF addresses this redundancy.