Search results

thanks @bitbckt I checked fs.com, seems to me that almost all of their switches are not ptp ready. That is pretty sad.

hi guys, I am looking for a used switch that is PTP IEEE 1588v2 ready, more specifically, it needs to be able to act as a PTP transparent clock so its processing & queuing delays can be included in PTP packets to be compensated. any model to recommend? thanks.

thanks @necr for the info. I checked all those relevant numbers and I do have lots of RX dropped/overruns. ifconfig output - This gives me 43Gb throughput when tested using iperf -P 10 using the default 1500 bytes MTU. When I changed MTU to 4000, I managed to get 50.5Gb throughput, so that...

I have two workstations with 40G/56GbE ConnectX-3 NICs connected to a Mellanox SX6012 switch, iperf and RoCE reported expected ~40Gb bandwidth when using 40Gb DAC cables. Everything was fine. Just picked up some 56Gb Mellanox DACs online, SX6012 recognized them as 56GbE cables and I had both...

Thanks for the info, does this mean PFC is required when starting to max out multiple ports? new to networking & RDMA, will really appreciate any help. :D

Thanks for the very informative replies above, learned tons of new stuff. Cheers!

@Necrotyr , thanks a lot for the info above. I am weighting my options to whether upgrade to ConnectX-3 Pro NICs (from ConnectX-3 NICs), the real benefit here is its RoCEv2 capability. I have managed to get RoCEv1 working properly in my setup and heard that if L3 routing is not required then...

Just picked up a Mellanox SX6012, it works in ethernet mode by default, probably the ex-owner configured it that way. Wondering does this switch support RoCE v2? I have some HP 544+FLR 40G NICs (Mellanox ConnectX-3 Pro chips which support RoCE v2) and planning to setup a RoCE testbed for homelab...

@BackupProphet Thanks a lot for your reply! I think I've found the issue - it was a bug in Linux Kernel 4.15. On Ubuntu 16.04 with its 4.4 kernel, everything seems to be fine: uname -a output: Once on Ubuntu 16.04 with its hwe kernel 4.15, fsync becomes stupidly slow: uname -a here...

I grabbed an Intel S3700 to use for SLOG kind of workload. I understand that write sync performance (fsync() latency) is the most important factor here. Did a quick pg_test_fsync run and found that fsync() latency is almost 8ms each write sync. This is very different from what I heard...

I have tried a different motherboard and another adaptor cable, both had the HDD LED being always solid when the system is idle. It does blinking on activity.

Thanks Rand for the reply. I am not worried about the HDD LED being lit/on all the time itself, more concerned on whether it indicates that the SSD is already faulty after less than 30 hours power on.

I recently bought an U.2 version of the Intel P3700 800G SSD. As my motherboard doesn't have U.2 support, I have to connect it to the M.2 slot on my motherboard using an U.2 to M.2 adaptor cable. Read/Write seem to be working fine for the new SSD, but the HDD LED on the front panel of my case...

It all come down to whether you know the market/sellers. I bought most of my home server stuff from taobao (2 x dual 2696v4, 1 x dual 2690v2), also 3 x 40GE mallanex cards, a 10g Huawei switch and almost a dozen SSDs (3 x 900P, 1 x 750, 4 x optane memory), never had any problem. I only buy...

ah, good to know that. Thanks! I also noticed that the pro version comes with "Ethernet connectivity with hardware offload engines to Overlay Networks (“Tunneling”)", which is not mentioned in the non-pro version. Is that something important...

I am in the market searching for a few 40G NICs, will mostly use them in EN mode. There are both Mellanox ConnectX 3 (MCX354A FCBT) and Mellanox ConnectX 3 Pro (MCX354A FCCT) available at roughly the same price. The pro version is obviously suppose to be better, wondering what is the actual...

I have a program continuously write data to the SSD at around 50-200kbytes per second. I am wondering whether such non-stop 24/7 continuous writes are going to significantly impact the life span of the SSDs. The concerned SSDs are enterprise class SSDs with pretty high write endurance rating...

sorry that I probably caused some confusion. the results I posted above was captured without any VM involved. I just have the feeling that if the VM gives you such a big performance hit that you end up with a NAND SSD class 4K Q1T1 performance, then there is probably something wrong somewhere...

the above 4K QD1T1 performance seems to be quite bad. 40-60Mbytes/sec is something I'd expect from a NAND SSD, you should get 200-300Mbytes/sec from your 900p. 4K Q32T1 is also not remotely close to what it should be. 280G 900p results found online.

I think a detailed P4800x vs 900p comparison with benchmarks will be very very interesting. So far, I already figured out from online pics that the controller is different, similar to the Intel 750 vs P3700 case, the optane chips are off by the last digit meaning the endurance ratings are...