A rack for supporting a sleds includes a pair of elongated support posts and pairs of elongated support arms that extend from the elongated support posts. Each pair of the elongated support arms defines a sled slot to receive a corresponding sled. To do so, each elongated support arm includes a circuit board guide to receive a chassis-less circuit board substrate of the corresponding sled. The rack may include a cross-member arm associated with each sled slot and an optical connector mounted to each cross-member arm. Additional elongated support posts may be used to provide additional sled slots.

Technologies for high-performance single-stream data compression include a computing device that updates an index data structure based on an input data stream. The input data stream is divided into multiple chunks. Each chunk has a predetermined length, such as 136 bytes, and overlaps the previous chunk by a predetermine amount, such as eight bytes. The computing device processes multiple chunks in parallel using the index data to generate multiple token streams. The tokens include literal tokens and reference tokens that refer to matching data from earlier in the input data stream. The computing device thus searches for matching data in parallel. The computing device merges the token streams to generate a single output token stream. The computing device may merge a pair of tokens from two different chunks to generate one or more synchronized tokens that are output to the output token stream. Other embodiments are described and claimed.

Adaptively compressing an input string (10) comprising a sequence of symbols in order to create a plurality of segment dictionaries D.sub.m, with the steps of: generating a lookup map (110); generating a key value segment S.sub.m,n; searching the lookup map for each symbol received in the input string (120, 130); upon detecting a symbol is not stored in the lookup map, adding the symbol by storing the symbol at a next sequential key index in the lookup map lookup map (135) and assigning a next sequential key value entry to the symbol and adding this key value to the key value segment S.sub.m,n (150); upon detecting the symbol is stored in the lookup map, adding the corresponding key value assigned to this symbol to the next sequential entry of the key value segment S.sub.m,n (150); wherein a new key value segment S.sub.m,n+1 of the lookup map is generated if the number of different symbols equals the number of available key values k=2.sup.n for the opened/current key value segment S.sub.m,n (141, 142), and where-in the lookup map is converted into a segment dictionary D.sub.m if the maximal key value size k.sub.nmax=2.sup.nmax is reached (132, 133, 134), with n being any positive integral number 1 to nmax, nmax denoting the maximal bit size, and m being any positive integral number denoting the consecutive numbering of segment dictionaries D.sub.m.

Technologies for blind mating of optical connectors in a rack of a data center are disclosed. In the illustrative embodiment, a sled can be slid into a rack and an optical connector on the sled will blindly mate with a corresponding optical connector on the rack. The illustrative optical connector on the sled includes two guide post receivers which mate with corresponding guide posts on the optical connector on the rack such that, when mated, optical fibers of the optical connector on the rack will be aligned and optically coupled with corresponding optical fibers on the optical connector of the sled.

Examples may include sleds for a rack in a data center including physical accelerator resources and memory for the accelerator resources. The memory can be shared between the accelerator resources. One or more memory controllers can be provided to couple the accelerator resources to the memory to provide memory access to all the accelerator resources. Each accelerator resource can include a memory controller to access a portion of the memory while the accelerator resources can be coupled via an out-of-band channel to provide memory access to the other portions of the memory.

Technologies for variable extent storage include multiple computing devices in communication over an optical fabric. A computing device receives a key-value storage request from an application that is indicative of a key. The computing device identifies one or more non-volatile storage blocks to store a value associated with the key and issues a non-volatile memory (NVM) input/output (I/O) command indicative of the NVM storage blocks to an NVM subsystem. The key-value storage request may include a read request or a store request, and the I/O command may include a read command or a write command. The I/O command may be issued to an NVM subsystem over the optical fabric. The computing device may be embodied as a storage sled of a data center, and the application may be executed by a compute sled of the data center. Other embodiments are described and claimed.

Technologies for low-latency compression in a data center are disclosed. In the illustrative embodiment, a storage sled compresses data with a low-latency compression algorithm prior to storing the data. The latency of the compression algorithm is less than the latency of the storage device, so that the latency of the storage and retrieval times are not significantly affected by the compression and decompression. In other embodiments, a compute sled may compress data with a low-latency compression algorithm prior to sending the data to a storage sled.

Technologies for enhanced memory wear leveling is disclosed. In the illustrative embodiment, a storage controller on a storage sled performs wear leveling across several storage devices. For example, the storage controller may copy hot data from one storage device that has a high number of erasures to another storage device that has a lower number of erasures in order to make the number of erasures between the devices more even by accumulating further erasures associated with the hot data on the drive that has the lower number of erasures.

Examples may include techniques to allocate physical accelerator resources from pools of accelerator resources. In particular, virtual computing devices can be composed from physical resources and physical accelerator resources dynamically allocated to the virtual computing devices. The present disclosure provides that physical accelerator resources can be dynamically allocated, or composed, to a virtual computing device despite not being physically coupled to other components in the virtual device.

Technologies for accelerating data writes include a managed node that includes a network interface controller that includes a power loss protected buffer and non-volatile memory. The managed node is to receive, through the network interface controller, a write request from a remote device. The write request includes a data block. The managed node is also to write the data block to the power loss protected buffer of the network interface controller, and send, in response to receipt of the data block and prior to a write of the data block to the non-volatile memory, an acknowledgement to the remote device. The acknowledgement is indicative of a successful write of the data block to the non-volatile memory. The managed node is also to write, after the acknowledgement has been sent, the data block from the power loss protected buffer to the non-volatile memory. Other embodiments are also described and claimed.